artificial intelligence dietician research paper

Artificial intelligence in nutrition research: perspectives on current and future applications

Affiliations.

• 1 Centre de recherche Nutrition, santé et société (NUTRISS), INAF, Université Laval, Québec, QC, Canada.
• 2 School of Nutrition, Université Laval, Québec, QC, Canada.
• PMID: 34525321
• DOI: 10.1139/apnm-2021-0448

Artificial intelligence (AI) is a rapidly evolving area that offers unparalleled opportunities of progress and applications in many healthcare fields. In this review, we provide an overview of the main and latest applications of AI in nutrition research and identify gaps to address to potentialize this emerging field. AI algorithms may help better understand and predict the complex and non-linear interactions between nutrition-related data and health outcomes, particularly when large amounts of data need to be structured and integrated, such as in metabolomics. AI-based approaches, including image recognition, may also improve dietary assessment by maximizing efficiency and addressing systematic and random errors associated with self-reported measurements of dietary intakes. Finally, AI applications can extract, structure and analyze large amounts of data from social media platforms to better understand dietary behaviours and perceptions among the population. In summary, AI-based approaches will likely improve and advance nutrition research as well as help explore new applications. However, further research is needed to identify areas where AI does deliver added value compared with traditional approaches, and other areas where AI is simply not likely to advance the field. Novelty: Artificial intelligence offers unparalleled opportunities of progress and applications in nutrition. There remain gaps to address to potentialize this emerging field.

Keywords: algorithmes; algorithms; apprentissage automatique; artificial intelligence; dietary assessment; intelligence artificielle; machine learning; metabolomics; médias sociaux; métabolomique; nutrition; prediction; prédiction; social media; évaluation alimentaire.

Artificial Intelligence (AI) based Nutrition Advisorusing an App

Ieee account.

• Change Username/Password
• Update Address

Purchase Details

• Payment Options
• Order History
• View Purchased Documents

Profile Information

• Communications Preferences
• Profession and Education
• Technical Interests
• US & Canada: +1 800 678 4333
• Worldwide: +1 732 981 0060
• Contact & Support
• About IEEE Xplore
• Accessibility
• Terms of Use
• Nondiscrimination Policy
• Privacy & Opting Out of Cookies

A not-for-profit organization, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity. © Copyright 2024 IEEE - All rights reserved. Use of this web site signifies your agreement to the terms and conditions.

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• My Account Login
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 26 March 2024

Predicting and improving complex beer flavor through machine learning

• Michiel Schreurs   ORCID: orcid.org/0000-0002-9449-5619 1 , 2 , 3   na1 ,
• Supinya Piampongsant 1 , 2 , 3   na1 ,
• Miguel Roncoroni   ORCID: orcid.org/0000-0001-7461-1427 1 , 2 , 3   na1 ,
• Lloyd Cool   ORCID: orcid.org/0000-0001-9936-3124 1 , 2 , 3 , 4 ,
• Beatriz Herrera-Malaver   ORCID: orcid.org/0000-0002-5096-9974 1 , 2 , 3 ,
• Christophe Vanderaa   ORCID: orcid.org/0000-0001-7443-5427 4 ,
• Florian A. Theßeling 1 , 2 , 3 ,
• Łukasz Kreft   ORCID: orcid.org/0000-0001-7620-4657 5 ,
• Alexander Botzki   ORCID: orcid.org/0000-0001-6691-4233 5 ,
• Philippe Malcorps 6 ,
• Luk Daenen 6 ,
• Tom Wenseleers   ORCID: orcid.org/0000-0002-1434-861X 4 &
• Kevin J. Verstrepen   ORCID: orcid.org/0000-0002-3077-6219 1 , 2 , 3  

Nature Communications volume  15 , Article number:  2368 ( 2024 ) Cite this article

41k Accesses

778 Altmetric

Metrics details

• Chemical engineering
• Gas chromatography
• Machine learning
• Metabolomics
• Taste receptors

The perception and appreciation of food flavor depends on many interacting chemical compounds and external factors, and therefore proves challenging to understand and predict. Here, we combine extensive chemical and sensory analyses of 250 different beers to train machine learning models that allow predicting flavor and consumer appreciation. For each beer, we measure over 200 chemical properties, perform quantitative descriptive sensory analysis with a trained tasting panel and map data from over 180,000 consumer reviews to train 10 different machine learning models. The best-performing algorithm, Gradient Boosting, yields models that significantly outperform predictions based on conventional statistics and accurately predict complex food features and consumer appreciation from chemical profiles. Model dissection allows identifying specific and unexpected compounds as drivers of beer flavor and appreciation. Adding these compounds results in variants of commercial alcoholic and non-alcoholic beers with improved consumer appreciation. Together, our study reveals how big data and machine learning uncover complex links between food chemistry, flavor and consumer perception, and lays the foundation to develop novel, tailored foods with superior flavors.

Similar content being viewed by others

BitterSweet: Building machine learning models for predicting the bitter and sweet taste of small molecules

Rudraksh Tuwani, Somin Wadhwa & Ganesh Bagler

Sensory lexicon and aroma volatiles analysis of brewing malt

Xiaoxia Su, Miao Yu, … Tianyi Du

Predicting odor from molecular structure: a multi-label classification approach

Kushagra Saini & Venkatnarayan Ramanathan

Introduction

Predicting and understanding food perception and appreciation is one of the major challenges in food science. Accurate modeling of food flavor and appreciation could yield important opportunities for both producers and consumers, including quality control, product fingerprinting, counterfeit detection, spoilage detection, and the development of new products and product combinations (food pairing) 1 , 2 , 3 , 4 , 5 , 6 . Accurate models for flavor and consumer appreciation would contribute greatly to our scientific understanding of how humans perceive and appreciate flavor. Moreover, accurate predictive models would also facilitate and standardize existing food assessment methods and could supplement or replace assessments by trained and consumer tasting panels, which are variable, expensive and time-consuming 7 , 8 , 9 . Lastly, apart from providing objective, quantitative, accurate and contextual information that can help producers, models can also guide consumers in understanding their personal preferences 10 .

Despite the myriad of applications, predicting food flavor and appreciation from its chemical properties remains a largely elusive goal in sensory science, especially for complex food and beverages 11 , 12 . A key obstacle is the immense number of flavor-active chemicals underlying food flavor. Flavor compounds can vary widely in chemical structure and concentration, making them technically challenging and labor-intensive to quantify, even in the face of innovations in metabolomics, such as non-targeted metabolic fingerprinting 13 , 14 . Moreover, sensory analysis is perhaps even more complicated. Flavor perception is highly complex, resulting from hundreds of different molecules interacting at the physiochemical and sensorial level. Sensory perception is often non-linear, characterized by complex and concentration-dependent synergistic and antagonistic effects 15 , 16 , 17 , 18 , 19 , 20 , 21 that are further convoluted by the genetics, environment, culture and psychology of consumers 22 , 23 , 24 . Perceived flavor is therefore difficult to measure, with problems of sensitivity, accuracy, and reproducibility that can only be resolved by gathering sufficiently large datasets 25 . Trained tasting panels are considered the prime source of quality sensory data, but require meticulous training, are low throughput and high cost. Public databases containing consumer reviews of food products could provide a valuable alternative, especially for studying appreciation scores, which do not require formal training 25 . Public databases offer the advantage of amassing large amounts of data, increasing the statistical power to identify potential drivers of appreciation. However, public datasets suffer from biases, including a bias in the volunteers that contribute to the database, as well as confounding factors such as price, cult status and psychological conformity towards previous ratings of the product.

Classical multivariate statistics and machine learning methods have been used to predict flavor of specific compounds by, for example, linking structural properties of a compound to its potential biological activities or linking concentrations of specific compounds to sensory profiles 1 , 26 . Importantly, most previous studies focused on predicting organoleptic properties of single compounds (often based on their chemical structure) 27 , 28 , 29 , 30 , 31 , 32 , 33 , thus ignoring the fact that these compounds are present in a complex matrix in food or beverages and excluding complex interactions between compounds. Moreover, the classical statistics commonly used in sensory science 34 , 35 , 36 , 37 , 38 , 39 require a large sample size and sufficient variance amongst predictors to create accurate models. They are not fit for studying an extensive set of hundreds of interacting flavor compounds, since they are sensitive to outliers, have a high tendency to overfit and are less suited for non-linear and discontinuous relationships 40 .

In this study, we combine extensive chemical analyses and sensory data of a set of different commercial beers with machine learning approaches to develop models that predict taste, smell, mouthfeel and appreciation from compound concentrations. Beer is particularly suited to model the relationship between chemistry, flavor and appreciation. First, beer is a complex product, consisting of thousands of flavor compounds that partake in complex sensory interactions 41 , 42 , 43 . This chemical diversity arises from the raw materials (malt, yeast, hops, water and spices) and biochemical conversions during the brewing process (kilning, mashing, boiling, fermentation, maturation and aging) 44 , 45 . Second, the advent of the internet saw beer consumers embrace online review platforms, such as RateBeer (ZX Ventures, Anheuser-Busch InBev SA/NV) and BeerAdvocate (Next Glass, inc.). In this way, the beer community provides massive data sets of beer flavor and appreciation scores, creating extraordinarily large sensory databases to complement the analyses of our professional sensory panel. Specifically, we characterize over 200 chemical properties of 250 commercial beers, spread across 22 beer styles, and link these to the descriptive sensory profiling data of a 16-person in-house trained tasting panel and data acquired from over 180,000 public consumer reviews. These unique and extensive datasets enable us to train a suite of machine learning models to predict flavor and appreciation from a beer’s chemical profile. Dissection of the best-performing models allows us to pinpoint specific compounds as potential drivers of beer flavor and appreciation. Follow-up experiments confirm the importance of these compounds and ultimately allow us to significantly improve the flavor and appreciation of selected commercial beers. Together, our study represents a significant step towards understanding complex flavors and reinforces the value of machine learning to develop and refine complex foods. In this way, it represents a stepping stone for further computer-aided food engineering applications 46 .

To generate a comprehensive dataset on beer flavor, we selected 250 commercial Belgian beers across 22 different beer styles (Supplementary Fig.  S1 ). Beers with ≤ 4.2% alcohol by volume (ABV) were classified as non-alcoholic and low-alcoholic. Blonds and Tripels constitute a significant portion of the dataset (12.4% and 11.2%, respectively) reflecting their presence on the Belgian beer market and the heterogeneity of beers within these styles. By contrast, lager beers are less diverse and dominated by a handful of brands. Rare styles such as Brut or Faro make up only a small fraction of the dataset (2% and 1%, respectively) because fewer of these beers are produced and because they are dominated by distinct characteristics in terms of flavor and chemical composition.

Extensive analysis identifies relationships between chemical compounds in beer

For each beer, we measured 226 different chemical properties, including common brewing parameters such as alcohol content, iso-alpha acids, pH, sugar concentration 47 , and over 200 flavor compounds (Methods, Supplementary Table  S1 ). A large portion (37.2%) are terpenoids arising from hopping, responsible for herbal and fruity flavors 16 , 48 . A second major category are yeast metabolites, such as esters and alcohols, that result in fruity and solvent notes 48 , 49 , 50 . Other measured compounds are primarily derived from malt, or other microbes such as non- Saccharomyces yeasts and bacteria (‘wild flora’). Compounds that arise from spices or staling are labeled under ‘Others’. Five attributes (caloric value, total acids and total ester, hop aroma and sulfur compounds) are calculated from multiple individually measured compounds.

As a first step in identifying relationships between chemical properties, we determined correlations between the concentrations of the compounds (Fig.  1 , upper panel, Supplementary Data  1 and 2 , and Supplementary Fig.  S2 . For the sake of clarity, only a subset of the measured compounds is shown in Fig.  1 ). Compounds of the same origin typically show a positive correlation, while absence of correlation hints at parameters varying independently. For example, the hop aroma compounds citronellol, and alpha-terpineol show moderate correlations with each other (Spearman’s rho=0.39 and 0.57), but not with the bittering hop component iso-alpha acids (Spearman’s rho=0.16 and −0.07). This illustrates how brewers can independently modify hop aroma and bitterness by selecting hop varieties and dosage time. If hops are added early in the boiling phase, chemical conversions increase bitterness while aromas evaporate, conversely, late addition of hops preserves aroma but limits bitterness 51 . Similarly, hop-derived iso-alpha acids show a strong anti-correlation with lactic acid and acetic acid, likely reflecting growth inhibition of lactic acid and acetic acid bacteria, or the consequent use of fewer hops in sour beer styles, such as West Flanders ales and Fruit beers, that rely on these bacteria for their distinct flavors 52 . Finally, yeast-derived esters (ethyl acetate, ethyl decanoate, ethyl hexanoate, ethyl octanoate) and alcohols (ethanol, isoamyl alcohol, isobutanol, and glycerol), correlate with Spearman coefficients above 0.5, suggesting that these secondary metabolites are correlated with the yeast genetic background and/or fermentation parameters and may be difficult to influence individually, although the choice of yeast strain may offer some control 53 .

Spearman rank correlations are shown. Descriptors are grouped according to their origin (malt (blue), hops (green), yeast (red), wild flora (yellow), Others (black)), and sensory aspect (aroma, taste, palate, and overall appreciation). Please note that for the chemical compounds, for the sake of clarity, only a subset of the total number of measured compounds is shown, with an emphasis on the key compounds for each source. For more details, see the main text and Methods section. Chemical data can be found in Supplementary Data  1 , correlations between all chemical compounds are depicted in Supplementary Fig.  S2 and correlation values can be found in Supplementary Data  2 . See Supplementary Data  4 for sensory panel assessments and Supplementary Data  5 for correlation values between all sensory descriptors.

Interestingly, different beer styles show distinct patterns for some flavor compounds (Supplementary Fig.  S3 ). These observations agree with expectations for key beer styles, and serve as a control for our measurements. For instance, Stouts generally show high values for color (darker), while hoppy beers contain elevated levels of iso-alpha acids, compounds associated with bitter hop taste. Acetic and lactic acid are not prevalent in most beers, with notable exceptions such as Kriek, Lambic, Faro, West Flanders ales and Flanders Old Brown, which use acid-producing bacteria ( Lactobacillus and Pediococcus ) or unconventional yeast ( Brettanomyces ) 54 , 55 . Glycerol, ethanol and esters show similar distributions across all beer styles, reflecting their common origin as products of yeast metabolism during fermentation 45 , 53 . Finally, low/no-alcohol beers contain low concentrations of glycerol and esters. This is in line with the production process for most of the low/no-alcohol beers in our dataset, which are produced through limiting fermentation or by stripping away alcohol via evaporation or dialysis, with both methods having the unintended side-effect of reducing the amount of flavor compounds in the final beer 56 , 57 .

Besides expected associations, our data also reveals less trivial associations between beer styles and specific parameters. For example, geraniol and citronellol, two monoterpenoids responsible for citrus, floral and rose flavors and characteristic of Citra hops, are found in relatively high amounts in Christmas, Saison, and Brett/co-fermented beers, where they may originate from terpenoid-rich spices such as coriander seeds instead of hops 58 .

Tasting panel assessments reveal sensorial relationships in beer

To assess the sensory profile of each beer, a trained tasting panel evaluated each of the 250 beers for 50 sensory attributes, including different hop, malt and yeast flavors, off-flavors and spices. Panelists used a tasting sheet (Supplementary Data  3 ) to score the different attributes. Panel consistency was evaluated by repeating 12 samples across different sessions and performing ANOVA. In 95% of cases no significant difference was found across sessions ( p  > 0.05), indicating good panel consistency (Supplementary Table  S2 ).

Aroma and taste perception reported by the trained panel are often linked (Fig.  1 , bottom left panel and Supplementary Data  4 and 5 ), with high correlations between hops aroma and taste (Spearman’s rho=0.83). Bitter taste was found to correlate with hop aroma and taste in general (Spearman’s rho=0.80 and 0.69), and particularly with “grassy” noble hops (Spearman’s rho=0.75). Barnyard flavor, most often associated with sour beers, is identified together with stale hops (Spearman’s rho=0.97) that are used in these beers. Lactic and acetic acid, which often co-occur, are correlated (Spearman’s rho=0.66). Interestingly, sweetness and bitterness are anti-correlated (Spearman’s rho = −0.48), confirming the hypothesis that they mask each other 59 , 60 . Beer body is highly correlated with alcohol (Spearman’s rho = 0.79), and overall appreciation is found to correlate with multiple aspects that describe beer mouthfeel (alcohol, carbonation; Spearman’s rho= 0.32, 0.39), as well as with hop and ester aroma intensity (Spearman’s rho=0.39 and 0.35).

Similar to the chemical analyses, sensorial analyses confirmed typical features of specific beer styles (Supplementary Fig.  S4 ). For example, sour beers (Faro, Flanders Old Brown, Fruit beer, Kriek, Lambic, West Flanders ale) were rated acidic, with flavors of both acetic and lactic acid. Hoppy beers were found to be bitter and showed hop-associated aromas like citrus and tropical fruit. Malt taste is most detected among scotch, stout/porters, and strong ales, while low/no-alcohol beers, which often have a reputation for being ‘worty’ (reminiscent of unfermented, sweet malt extract) appear in the middle. Unsurprisingly, hop aromas are most strongly detected among hoppy beers. Like its chemical counterpart (Supplementary Fig.  S3 ), acidity shows a right-skewed distribution, with the most acidic beers being Krieks, Lambics, and West Flanders ales.

Tasting panel assessments of specific flavors correlate with chemical composition

We find that the concentrations of several chemical compounds strongly correlate with specific aroma or taste, as evaluated by the tasting panel (Fig.  2 , Supplementary Fig.  S5 , Supplementary Data  6 ). In some cases, these correlations confirm expectations and serve as a useful control for data quality. For example, iso-alpha acids, the bittering compounds in hops, strongly correlate with bitterness (Spearman’s rho=0.68), while ethanol and glycerol correlate with tasters’ perceptions of alcohol and body, the mouthfeel sensation of fullness (Spearman’s rho=0.82/0.62 and 0.72/0.57 respectively) and darker color from roasted malts is a good indication of malt perception (Spearman’s rho=0.54).

Heatmap colors indicate Spearman’s Rho. Axes are organized according to sensory categories (aroma, taste, mouthfeel, overall), chemical categories and chemical sources in beer (malt (blue), hops (green), yeast (red), wild flora (yellow), Others (black)). See Supplementary Data  6 for all correlation values.

Interestingly, for some relationships between chemical compounds and perceived flavor, correlations are weaker than expected. For example, the rose-smelling phenethyl acetate only weakly correlates with floral aroma. This hints at more complex relationships and interactions between compounds and suggests a need for a more complex model than simple correlations. Lastly, we uncovered unexpected correlations. For instance, the esters ethyl decanoate and ethyl octanoate appear to correlate slightly with hop perception and bitterness, possibly due to their fruity flavor. Iron is anti-correlated with hop aromas and bitterness, most likely because it is also anti-correlated with iso-alpha acids. This could be a sign of metal chelation of hop acids 61 , given that our analyses measure unbound hop acids and total iron content, or could result from the higher iron content in dark and Fruit beers, which typically have less hoppy and bitter flavors 62 .

Public consumer reviews complement expert panel data

To complement and expand the sensory data of our trained tasting panel, we collected 180,000 reviews of our 250 beers from the online consumer review platform RateBeer. This provided numerical scores for beer appearance, aroma, taste, palate, overall quality as well as the average overall score.

Public datasets are known to suffer from biases, such as price, cult status and psychological conformity towards previous ratings of a product. For example, prices correlate with appreciation scores for these online consumer reviews (rho=0.49, Supplementary Fig.  S6 ), but not for our trained tasting panel (rho=0.19). This suggests that prices affect consumer appreciation, which has been reported in wine 63 , while blind tastings are unaffected. Moreover, we observe that some beer styles, like lagers and non-alcoholic beers, generally receive lower scores, reflecting that online reviewers are mostly beer aficionados with a preference for specialty beers over lager beers. In general, we find a modest correlation between our trained panel’s overall appreciation score and the online consumer appreciation scores (Fig.  3 , rho=0.29). Apart from the aforementioned biases in the online datasets, serving temperature, sample freshness and surroundings, which are all tightly controlled during the tasting panel sessions, can vary tremendously across online consumers and can further contribute to (among others, appreciation) differences between the two categories of tasters. Importantly, in contrast to the overall appreciation scores, for many sensory aspects the results from the professional panel correlated well with results obtained from RateBeer reviews. Correlations were highest for features that are relatively easy to recognize even for untrained tasters, like bitterness, sweetness, alcohol and malt aroma (Fig.  3 and below).

RateBeer text mining results can be found in Supplementary Data  7 . Rho values shown are Spearman correlation values, with asterisks indicating significant correlations ( p  < 0.05, two-sided). All p values were smaller than 0.001, except for Esters aroma (0.0553), Esters taste (0.3275), Esters aroma—banana (0.0019), Coriander (0.0508) and Diacetyl (0.0134).

Besides collecting consumer appreciation from these online reviews, we developed automated text analysis tools to gather additional data from review texts (Supplementary Data  7 ). Processing review texts on the RateBeer database yielded comparable results to the scores given by the trained panel for many common sensory aspects, including acidity, bitterness, sweetness, alcohol, malt, and hop tastes (Fig.  3 ). This is in line with what would be expected, since these attributes require less training for accurate assessment and are less influenced by environmental factors such as temperature, serving glass and odors in the environment. Consumer reviews also correlate well with our trained panel for 4-vinyl guaiacol, a compound associated with a very characteristic aroma. By contrast, correlations for more specific aromas like ester, coriander or diacetyl are underrepresented in the online reviews, underscoring the importance of using a trained tasting panel and standardized tasting sheets with explicit factors to be scored for evaluating specific aspects of a beer. Taken together, our results suggest that public reviews are trustworthy for some, but not all, flavor features and can complement or substitute taste panel data for these sensory aspects.

Models can predict beer sensory profiles from chemical data

The rich datasets of chemical analyses, tasting panel assessments and public reviews gathered in the first part of this study provided us with a unique opportunity to develop predictive models that link chemical data to sensorial features. Given the complexity of beer flavor, basic statistical tools such as correlations or linear regression may not always be the most suitable for making accurate predictions. Instead, we applied different machine learning models that can model both simple linear and complex interactive relationships. Specifically, we constructed a set of regression models to predict (a) trained panel scores for beer flavor and quality and (b) public reviews’ appreciation scores from beer chemical profiles. We trained and tested 10 different models (Methods), 3 linear regression-based models (simple linear regression with first-order interactions (LR), lasso regression with first-order interactions (Lasso), partial least squares regressor (PLSR)), 5 decision tree models (AdaBoost regressor (ABR), extra trees (ET), gradient boosting regressor (GBR), random forest (RF) and XGBoost regressor (XGBR)), 1 support vector regression (SVR), and 1 artificial neural network (ANN) model.

To compare the performance of our machine learning models, the dataset was randomly split into a training and test set, stratified by beer style. After a model was trained on data in the training set, its performance was evaluated on its ability to predict the test dataset obtained from multi-output models (based on the coefficient of determination, see Methods). Additionally, individual-attribute models were ranked per descriptor and the average rank was calculated, as proposed by Korneva et al. 64 . Importantly, both ways of evaluating the models’ performance agreed in general. Performance of the different models varied (Table  1 ). It should be noted that all models perform better at predicting RateBeer results than results from our trained tasting panel. One reason could be that sensory data is inherently variable, and this variability is averaged out with the large number of public reviews from RateBeer. Additionally, all tree-based models perform better at predicting taste than aroma. Linear models (LR) performed particularly poorly, with negative R 2 values, due to severe overfitting (training set R 2  = 1). Overfitting is a common issue in linear models with many parameters and limited samples, especially with interaction terms further amplifying the number of parameters. L1 regularization (Lasso) successfully overcomes this overfitting, out-competing multiple tree-based models on the RateBeer dataset. Similarly, the dimensionality reduction of PLSR avoids overfitting and improves performance, to some extent. Still, tree-based models (ABR, ET, GBR, RF and XGBR) show the best performance, out-competing the linear models (LR, Lasso, PLSR) commonly used in sensory science 65 .

GBR models showed the best overall performance in predicting sensory responses from chemical information, with R 2 values up to 0.75 depending on the predicted sensory feature (Supplementary Table  S4 ). The GBR models predict consumer appreciation (RateBeer) better than our trained panel’s appreciation (R 2 value of 0.67 compared to R 2 value of 0.09) (Supplementary Table  S3 and Supplementary Table  S4 ). ANN models showed intermediate performance, likely because neural networks typically perform best with larger datasets 66 . The SVR shows intermediate performance, mostly due to the weak predictions of specific attributes that lower the overall performance (Supplementary Table  S4 ).

Model dissection identifies specific, unexpected compounds as drivers of consumer appreciation

Next, we leveraged our models to infer important contributors to sensory perception and consumer appreciation. Consumer preference is a crucial sensory aspects, because a product that shows low consumer appreciation scores often does not succeed commercially 25 . Additionally, the requirement for a large number of representative evaluators makes consumer trials one of the more costly and time-consuming aspects of product development. Hence, a model for predicting chemical drivers of overall appreciation would be a welcome addition to the available toolbox for food development and optimization.

Since GBR models on our RateBeer dataset showed the best overall performance, we focused on these models. Specifically, we used two approaches to identify important contributors. First, rankings of the most important predictors for each sensorial trait in the GBR models were obtained based on impurity-based feature importance (mean decrease in impurity). High-ranked parameters were hypothesized to be either the true causal chemical properties underlying the trait, to correlate with the actual causal properties, or to take part in sensory interactions affecting the trait 67 (Fig.  4A ). In a second approach, we used SHAP 68 to determine which parameters contributed most to the model for making predictions of consumer appreciation (Fig.  4B ). SHAP calculates parameter contributions to model predictions on a per-sample basis, which can be aggregated into an importance score.

A The impurity-based feature importance (mean deviance in impurity, MDI) calculated from the Gradient Boosting Regression (GBR) model predicting RateBeer appreciation scores. The top 15 highest ranked chemical properties are shown. B SHAP summary plot for the top 15 parameters contributing to our GBR model. Each point on the graph represents a sample from our dataset. The color represents the concentration of that parameter, with bluer colors representing low values and redder colors representing higher values. Greater absolute values on the horizontal axis indicate a higher impact of the parameter on the prediction of the model. C Spearman correlations between the 15 most important chemical properties and consumer overall appreciation. Numbers indicate the Spearman Rho correlation coefficient, and the rank of this correlation compared to all other correlations. The top 15 important compounds were determined using SHAP (panel B).

Both approaches identified ethyl acetate as the most predictive parameter for beer appreciation (Fig.  4 ). Ethyl acetate is the most abundant ester in beer with a typical ‘fruity’, ‘solvent’ and ‘alcoholic’ flavor, but is often considered less important than other esters like isoamyl acetate. The second most important parameter identified by SHAP is ethanol, the most abundant beer compound after water. Apart from directly contributing to beer flavor and mouthfeel, ethanol drastically influences the physical properties of beer, dictating how easily volatile compounds escape the beer matrix to contribute to beer aroma 69 . Importantly, it should also be noted that the importance of ethanol for appreciation is likely inflated by the very low appreciation scores of non-alcoholic beers (Supplementary Fig.  S4 ). Despite not often being considered a driver of beer appreciation, protein level also ranks highly in both approaches, possibly due to its effect on mouthfeel and body 70 . Lactic acid, which contributes to the tart taste of sour beers, is the fourth most important parameter identified by SHAP, possibly due to the generally high appreciation of sour beers in our dataset.

Interestingly, some of the most important predictive parameters for our model are not well-established as beer flavors or are even commonly regarded as being negative for beer quality. For example, our models identify methanethiol and ethyl phenyl acetate, an ester commonly linked to beer staling 71 , as a key factor contributing to beer appreciation. Although there is no doubt that high concentrations of these compounds are considered unpleasant, the positive effects of modest concentrations are not yet known 72 , 73 .

To compare our approach to conventional statistics, we evaluated how well the 15 most important SHAP-derived parameters correlate with consumer appreciation (Fig.  4C ). Interestingly, only 6 of the properties derived by SHAP rank amongst the top 15 most correlated parameters. For some chemical compounds, the correlations are so low that they would have likely been considered unimportant. For example, lactic acid, the fourth most important parameter, shows a bimodal distribution for appreciation, with sour beers forming a separate cluster, that is missed entirely by the Spearman correlation. Additionally, the correlation plots reveal outliers, emphasizing the need for robust analysis tools. Together, this highlights the need for alternative models, like the Gradient Boosting model, that better grasp the complexity of (beer) flavor.

Finally, to observe the relationships between these chemical properties and their predicted targets, partial dependence plots were constructed for the six most important predictors of consumer appreciation 74 , 75 , 76 (Supplementary Fig.  S7 ). One-way partial dependence plots show how a change in concentration affects the predicted appreciation. These plots reveal an important limitation of our models: appreciation predictions remain constant at ever-increasing concentrations. This implies that once a threshold concentration is reached, further increasing the concentration does not affect appreciation. This is false, as it is well-documented that certain compounds become unpleasant at high concentrations, including ethyl acetate (‘nail polish’) 77 and methanethiol (‘sulfury’ and ‘rotten cabbage’) 78 . The inability of our models to grasp that flavor compounds have optimal levels, above which they become negative, is a consequence of working with commercial beer brands where (off-)flavors are rarely too high to negatively impact the product. The two-way partial dependence plots show how changing the concentration of two compounds influences predicted appreciation, visualizing their interactions (Supplementary Fig.  S7 ). In our case, the top 5 parameters are dominated by additive or synergistic interactions, with high concentrations for both compounds resulting in the highest predicted appreciation.

To assess the robustness of our best-performing models and model predictions, we performed 100 iterations of the GBR, RF and ET models. In general, all iterations of the models yielded similar performance (Supplementary Fig.  S8 ). Moreover, the main predictors (including the top predictors ethanol and ethyl acetate) remained virtually the same, especially for GBR and RF. For the iterations of the ET model, we did observe more variation in the top predictors, which is likely a consequence of the model’s inherent random architecture in combination with co-correlations between certain predictors. However, even in this case, several of the top predictors (ethanol and ethyl acetate) remain unchanged, although their rank in importance changes (Supplementary Fig.  S8 ).

Next, we investigated if a combination of RateBeer and trained panel data into one consolidated dataset would lead to stronger models, under the hypothesis that such a model would suffer less from bias in the datasets. A GBR model was trained to predict appreciation on the combined dataset. This model underperformed compared to the RateBeer model, both in the native case and when including a dataset identifier (R 2  = 0.67, 0.26 and 0.42 respectively). For the latter, the dataset identifier is the most important feature (Supplementary Fig.  S9 ), while most of the feature importance remains unchanged, with ethyl acetate and ethanol ranking highest, like in the original model trained only on RateBeer data. It seems that the large variation in the panel dataset introduces noise, weakening the models’ performances and reliability. In addition, it seems reasonable to assume that both datasets are fundamentally different, with the panel dataset obtained by blind tastings by a trained professional panel.

Lastly, we evaluated whether beer style identifiers would further enhance the model’s performance. A GBR model was trained with parameters that explicitly encoded the styles of the samples. This did not improve model performance (R2 = 0.66 with style information vs R2 = 0.67). The most important chemical features are consistent with the model trained without style information (eg. ethanol and ethyl acetate), and with the exception of the most preferred (strong ale) and least preferred (low/no-alcohol) styles, none of the styles were among the most important features (Supplementary Fig.  S9 , Supplementary Table  S5 and S6 ). This is likely due to a combination of style-specific chemical signatures, such as iso-alpha acids and lactic acid, that implicitly convey style information to the original models, as well as the low number of samples belonging to some styles, making it difficult for the model to learn style-specific patterns. Moreover, beer styles are not rigorously defined, with some styles overlapping in features and some beers being misattributed to a specific style, all of which leads to more noise in models that use style parameters.

Model validation

To test if our predictive models give insight into beer appreciation, we set up experiments aimed at improving existing commercial beers. We specifically selected overall appreciation as the trait to be examined because of its complexity and commercial relevance. Beer flavor comprises a complex bouquet rather than single aromas and tastes 53 . Hence, adding a single compound to the extent that a difference is noticeable may lead to an unbalanced, artificial flavor. Therefore, we evaluated the effect of combinations of compounds. Because Blond beers represent the most extensive style in our dataset, we selected a beer from this style as the starting material for these experiments (Beer 64 in Supplementary Data  1 ).

In the first set of experiments, we adjusted the concentrations of compounds that made up the most important predictors of overall appreciation (ethyl acetate, ethanol, lactic acid, ethyl phenyl acetate) together with correlated compounds (ethyl hexanoate, isoamyl acetate, glycerol), bringing them up to 95 th percentile ethanol-normalized concentrations (Methods) within the Blond group (‘Spiked’ concentration in Fig.  5A ). Compared to controls, the spiked beers were found to have significantly improved overall appreciation among trained panelists, with panelist noting increased intensity of ester flavors, sweetness, alcohol, and body fullness (Fig.  5B ). To disentangle the contribution of ethanol to these results, a second experiment was performed without the addition of ethanol. This resulted in a similar outcome, including increased perception of alcohol and overall appreciation.

Adding the top chemical compounds, identified as best predictors of appreciation by our model, into poorly appreciated beers results in increased appreciation from our trained panel. Results of sensory tests between base beers and those spiked with compounds identified as the best predictors by the model. A Blond and Non/Low-alcohol (0.0% ABV) base beers were brought up to 95th-percentile ethanol-normalized concentrations within each style. B For each sensory attribute, tasters indicated the more intense sample and selected the sample they preferred. The numbers above the bars correspond to the p values that indicate significant changes in perceived flavor (two-sided binomial test: alpha 0.05, n  = 20 or 13).

In a last experiment, we tested whether using the model’s predictions can boost the appreciation of a non-alcoholic beer (beer 223 in Supplementary Data  1 ). Again, the addition of a mixture of predicted compounds (omitting ethanol, in this case) resulted in a significant increase in appreciation, body, ester flavor and sweetness.

Predicting flavor and consumer appreciation from chemical composition is one of the ultimate goals of sensory science. A reliable, systematic and unbiased way to link chemical profiles to flavor and food appreciation would be a significant asset to the food and beverage industry. Such tools would substantially aid in quality control and recipe development, offer an efficient and cost-effective alternative to pilot studies and consumer trials and would ultimately allow food manufacturers to produce superior, tailor-made products that better meet the demands of specific consumer groups more efficiently.

A limited set of studies have previously tried, to varying degrees of success, to predict beer flavor and beer popularity based on (a limited set of) chemical compounds and flavors 79 , 80 . Current sensitive, high-throughput technologies allow measuring an unprecedented number of chemical compounds and properties in a large set of samples, yielding a dataset that can train models that help close the gaps between chemistry and flavor, even for a complex natural product like beer. To our knowledge, no previous research gathered data at this scale (250 samples, 226 chemical parameters, 50 sensory attributes and 5 consumer scores) to disentangle and validate the chemical aspects driving beer preference using various machine-learning techniques. We find that modern machine learning models outperform conventional statistical tools, such as correlations and linear models, and can successfully predict flavor appreciation from chemical composition. This could be attributed to the natural incorporation of interactions and non-linear or discontinuous effects in machine learning models, which are not easily grasped by the linear model architecture. While linear models and partial least squares regression represent the most widespread statistical approaches in sensory science, in part because they allow interpretation 65 , 81 , 82 , modern machine learning methods allow for building better predictive models while preserving the possibility to dissect and exploit the underlying patterns. Of the 10 different models we trained, tree-based models, such as our best performing GBR, showed the best overall performance in predicting sensory responses from chemical information, outcompeting artificial neural networks. This agrees with previous reports for models trained on tabular data 83 . Our results are in line with the findings of Colantonio et al. who also identified the gradient boosting architecture as performing best at predicting appreciation and flavor (of tomatoes and blueberries, in their specific study) 26 . Importantly, besides our larger experimental scale, we were able to directly confirm our models’ predictions in vivo.

Our study confirms that flavor compound concentration does not always correlate with perception, suggesting complex interactions that are often missed by more conventional statistics and simple models. Specifically, we find that tree-based algorithms may perform best in developing models that link complex food chemistry with aroma. Furthermore, we show that massive datasets of untrained consumer reviews provide a valuable source of data, that can complement or even replace trained tasting panels, especially for appreciation and basic flavors, such as sweetness and bitterness. This holds despite biases that are known to occur in such datasets, such as price or conformity bias. Moreover, GBR models predict taste better than aroma. This is likely because taste (e.g. bitterness) often directly relates to the corresponding chemical measurements (e.g., iso-alpha acids), whereas such a link is less clear for aromas, which often result from the interplay between multiple volatile compounds. We also find that our models are best at predicting acidity and alcohol, likely because there is a direct relation between the measured chemical compounds (acids and ethanol) and the corresponding perceived sensorial attribute (acidity and alcohol), and because even untrained consumers are generally able to recognize these flavors and aromas.

The predictions of our final models, trained on review data, hold even for blind tastings with small groups of trained tasters, as demonstrated by our ability to validate specific compounds as drivers of beer flavor and appreciation. Since adding a single compound to the extent of a noticeable difference may result in an unbalanced flavor profile, we specifically tested our identified key drivers as a combination of compounds. While this approach does not allow us to validate if a particular single compound would affect flavor and/or appreciation, our experiments do show that this combination of compounds increases consumer appreciation.

It is important to stress that, while it represents an important step forward, our approach still has several major limitations. A key weakness of the GBR model architecture is that amongst co-correlating variables, the largest main effect is consistently preferred for model building. As a result, co-correlating variables often have artificially low importance scores, both for impurity and SHAP-based methods, like we observed in the comparison to the more randomized Extra Trees models. This implies that chemicals identified as key drivers of a specific sensory feature by GBR might not be the true causative compounds, but rather co-correlate with the actual causative chemical. For example, the high importance of ethyl acetate could be (partially) attributed to the total ester content, ethanol or ethyl hexanoate (rho=0.77, rho=0.72 and rho=0.68), while ethyl phenylacetate could hide the importance of prenyl isobutyrate and ethyl benzoate (rho=0.77 and rho=0.76). Expanding our GBR model to include beer style as a parameter did not yield additional power or insight. This is likely due to style-specific chemical signatures, such as iso-alpha acids and lactic acid, that implicitly convey style information to the original model, as well as the smaller sample size per style, limiting the power to uncover style-specific patterns. This can be partly attributed to the curse of dimensionality, where the high number of parameters results in the models mainly incorporating single parameter effects, rather than complex interactions such as style-dependent effects 67 . A larger number of samples may overcome some of these limitations and offer more insight into style-specific effects. On the other hand, beer style is not a rigid scientific classification, and beers within one style often differ a lot, which further complicates the analysis of style as a model factor.

Our study is limited to beers from Belgian breweries. Although these beers cover a large portion of the beer styles available globally, some beer styles and consumer patterns may be missing, while other features might be overrepresented. For example, many Belgian ales exhibit yeast-driven flavor profiles, which is reflected in the chemical drivers of appreciation discovered by this study. In future work, expanding the scope to include diverse markets and beer styles could lead to the identification of even more drivers of appreciation and better models for special niche products that were not present in our beer set.

In addition to inherent limitations of GBR models, there are also some limitations associated with studying food aroma. Even if our chemical analyses measured most of the known aroma compounds, the total number of flavor compounds in complex foods like beer is still larger than the subset we were able to measure in this study. For example, hop-derived thiols, that influence flavor at very low concentrations, are notoriously difficult to measure in a high-throughput experiment. Moreover, consumer perception remains subjective and prone to biases that are difficult to avoid. It is also important to stress that the models are still immature and that more extensive datasets will be crucial for developing more complete models in the future. Besides more samples and parameters, our dataset does not include any demographic information about the tasters. Including such data could lead to better models that grasp external factors like age and culture. Another limitation is that our set of beers consists of high-quality end-products and lacks beers that are unfit for sale, which limits the current model in accurately predicting products that are appreciated very badly. Finally, while models could be readily applied in quality control, their use in sensory science and product development is restrained by their inability to discern causal relationships. Given that the models cannot distinguish compounds that genuinely drive consumer perception from those that merely correlate, validation experiments are essential to identify true causative compounds.

Despite the inherent limitations, dissection of our models enabled us to pinpoint specific molecules as potential drivers of beer aroma and consumer appreciation, including compounds that were unexpected and would not have been identified using standard approaches. Important drivers of beer appreciation uncovered by our models include protein levels, ethyl acetate, ethyl phenyl acetate and lactic acid. Currently, many brewers already use lactic acid to acidify their brewing water and ensure optimal pH for enzymatic activity during the mashing process. Our results suggest that adding lactic acid can also improve beer appreciation, although its individual effect remains to be tested. Interestingly, ethanol appears to be unnecessary to improve beer appreciation, both for blond beer and alcohol-free beer. Given the growing consumer interest in alcohol-free beer, with a predicted annual market growth of >7% 84 , it is relevant for brewers to know what compounds can further increase consumer appreciation of these beers. Hence, our model may readily provide avenues to further improve the flavor and consumer appreciation of both alcoholic and non-alcoholic beers, which is generally considered one of the key challenges for future beer production.

Whereas we see a direct implementation of our results for the development of superior alcohol-free beverages and other food products, our study can also serve as a stepping stone for the development of novel alcohol-containing beverages. We want to echo the growing body of scientific evidence for the negative effects of alcohol consumption, both on the individual level by the mutagenic, teratogenic and carcinogenic effects of ethanol 85 , 86 , as well as the burden on society caused by alcohol abuse and addiction. We encourage the use of our results for the production of healthier, tastier products, including novel and improved beverages with lower alcohol contents. Furthermore, we strongly discourage the use of these technologies to improve the appreciation or addictive properties of harmful substances.

The present work demonstrates that despite some important remaining hurdles, combining the latest developments in chemical analyses, sensory analysis and modern machine learning methods offers exciting avenues for food chemistry and engineering. Soon, these tools may provide solutions in quality control and recipe development, as well as new approaches to sensory science and flavor research.

Beer selection

250 commercial Belgian beers were selected to cover the broad diversity of beer styles and corresponding diversity in chemical composition and aroma. See Supplementary Fig.  S1 .

Chemical dataset

Sample preparation.

Beers within their expiration date were purchased from commercial retailers. Samples were prepared in biological duplicates at room temperature, unless explicitly stated otherwise. Bottle pressure was measured with a manual pressure device (Steinfurth Mess-Systeme GmbH) and used to calculate CO 2 concentration. The beer was poured through two filter papers (Macherey-Nagel, 500713032 MN 713 ¼) to remove carbon dioxide and prevent spontaneous foaming. Samples were then prepared for measurements by targeted Headspace-Gas Chromatography-Flame Ionization Detector/Flame Photometric Detector (HS-GC-FID/FPD), Headspace-Solid Phase Microextraction-Gas Chromatography-Mass Spectrometry (HS-SPME-GC-MS), colorimetric analysis, enzymatic analysis, Near-Infrared (NIR) analysis, as described in the sections below. The mean values of biological duplicates are reported for each compound.

HS-GC-FID/FPD

HS-GC-FID/FPD (Shimadzu GC 2010 Plus) was used to measure higher alcohols, acetaldehyde, esters, 4-vinyl guaicol, and sulfur compounds. Each measurement comprised 5 ml of sample pipetted into a 20 ml glass vial containing 1.75 g NaCl (VWR, 27810.295). 100 µl of 2-heptanol (Sigma-Aldrich, H3003) (internal standard) solution in ethanol (Fisher Chemical, E/0650DF/C17) was added for a final concentration of 2.44 mg/L. Samples were flushed with nitrogen for 10 s, sealed with a silicone septum, stored at −80 °C and analyzed in batches of 20.

The GC was equipped with a DB-WAXetr column (length, 30 m; internal diameter, 0.32 mm; layer thickness, 0.50 µm; Agilent Technologies, Santa Clara, CA, USA) to the FID and an HP-5 column (length, 30 m; internal diameter, 0.25 mm; layer thickness, 0.25 µm; Agilent Technologies, Santa Clara, CA, USA) to the FPD. N 2 was used as the carrier gas. Samples were incubated for 20 min at 70 °C in the headspace autosampler (Flow rate, 35 cm/s; Injection volume, 1000 µL; Injection mode, split; Combi PAL autosampler, CTC analytics, Switzerland). The injector, FID and FPD temperatures were kept at 250 °C. The GC oven temperature was first held at 50 °C for 5 min and then allowed to rise to 80 °C at a rate of 5 °C/min, followed by a second ramp of 4 °C/min until 200 °C kept for 3 min and a final ramp of (4 °C/min) until 230 °C for 1 min. Results were analyzed with the GCSolution software version 2.4 (Shimadzu, Kyoto, Japan). The GC was calibrated with a 5% EtOH solution (VWR International) containing the volatiles under study (Supplementary Table  S7 ).

HS-SPME-GC-MS

HS-SPME-GC-MS (Shimadzu GCMS-QP-2010 Ultra) was used to measure additional volatile compounds, mainly comprising terpenoids and esters. Samples were analyzed by HS-SPME using a triphase DVB/Carboxen/PDMS 50/30 μm SPME fiber (Supelco Co., Bellefonte, PA, USA) followed by gas chromatography (Thermo Fisher Scientific Trace 1300 series, USA) coupled to a mass spectrometer (Thermo Fisher Scientific ISQ series MS) equipped with a TriPlus RSH autosampler. 5 ml of degassed beer sample was placed in 20 ml vials containing 1.75 g NaCl (VWR, 27810.295). 5 µl internal standard mix was added, containing 2-heptanol (1 g/L) (Sigma-Aldrich, H3003), 4-fluorobenzaldehyde (1 g/L) (Sigma-Aldrich, 128376), 2,3-hexanedione (1 g/L) (Sigma-Aldrich, 144169) and guaiacol (1 g/L) (Sigma-Aldrich, W253200) in ethanol (Fisher Chemical, E/0650DF/C17). Each sample was incubated at 60 °C in the autosampler oven with constant agitation. After 5 min equilibration, the SPME fiber was exposed to the sample headspace for 30 min. The compounds trapped on the fiber were thermally desorbed in the injection port of the chromatograph by heating the fiber for 15 min at 270 °C.

The GC-MS was equipped with a low polarity RXi-5Sil MS column (length, 20 m; internal diameter, 0.18 mm; layer thickness, 0.18 µm; Restek, Bellefonte, PA, USA). Injection was performed in splitless mode at 320 °C, a split flow of 9 ml/min, a purge flow of 5 ml/min and an open valve time of 3 min. To obtain a pulsed injection, a programmed gas flow was used whereby the helium gas flow was set at 2.7 mL/min for 0.1 min, followed by a decrease in flow of 20 ml/min to the normal 0.9 mL/min. The temperature was first held at 30 °C for 3 min and then allowed to rise to 80 °C at a rate of 7 °C/min, followed by a second ramp of 2 °C/min till 125 °C and a final ramp of 8 °C/min with a final temperature of 270 °C.

Mass acquisition range was 33 to 550 amu at a scan rate of 5 scans/s. Electron impact ionization energy was 70 eV. The interface and ion source were kept at 275 °C and 250 °C, respectively. A mix of linear n-alkanes (from C7 to C40, Supelco Co.) was injected into the GC-MS under identical conditions to serve as external retention index markers. Identification and quantification of the compounds were performed using an in-house developed R script as described in Goelen et al. and Reher et al. 87 , 88 (for package information, see Supplementary Table  S8 ). Briefly, chromatograms were analyzed using AMDIS (v2.71) 89 to separate overlapping peaks and obtain pure compound spectra. The NIST MS Search software (v2.0 g) in combination with the NIST2017, FFNSC3 and Adams4 libraries were used to manually identify the empirical spectra, taking into account the expected retention time. After background subtraction and correcting for retention time shifts between samples run on different days based on alkane ladders, compound elution profiles were extracted and integrated using a file with 284 target compounds of interest, which were either recovered in our identified AMDIS list of spectra or were known to occur in beer. Compound elution profiles were estimated for every peak in every chromatogram over a time-restricted window using weighted non-negative least square analysis after which peak areas were integrated 87 , 88 . Batch effect correction was performed by normalizing against the most stable internal standard compound, 4-fluorobenzaldehyde. Out of all 284 target compounds that were analyzed, 167 were visually judged to have reliable elution profiles and were used for final analysis.

Discrete photometric and enzymatic analysis

Discrete photometric and enzymatic analysis (Thermo Scientific TM Gallery TM Plus Beermaster Discrete Analyzer) was used to measure acetic acid, ammonia, beta-glucan, iso-alpha acids, color, sugars, glycerol, iron, pH, protein, and sulfite. 2 ml of sample volume was used for the analyses. Information regarding the reagents and standard solutions used for analyses and calibrations is included in Supplementary Table  S7 and Supplementary Table  S9 .

NIR analyses

NIR analysis (Anton Paar Alcolyzer Beer ME System) was used to measure ethanol. Measurements comprised 50 ml of sample, and a 10% EtOH solution was used for calibration.

Correlation calculations

Pairwise Spearman Rank correlations were calculated between all chemical properties.

Sensory dataset

Trained panel.

Our trained tasting panel consisted of volunteers who gave prior verbal informed consent. All compounds used for the validation experiment were of food-grade quality. The tasting sessions were approved by the Social and Societal Ethics Committee of the KU Leuven (G-2022-5677-R2(MAR)). All online reviewers agreed to the Terms and Conditions of the RateBeer website.

Sensory analysis was performed according to the American Society of Brewing Chemists (ASBC) Sensory Analysis Methods 90 . 30 volunteers were screened through a series of triangle tests. The sixteen most sensitive and consistent tasters were retained as taste panel members. The resulting panel was diverse in age [22–42, mean: 29], sex [56% male] and nationality [7 different countries]. The panel developed a consensus vocabulary to describe beer aroma, taste and mouthfeel. Panelists were trained to identify and score 50 different attributes, using a 7-point scale to rate attributes’ intensity. The scoring sheet is included as Supplementary Data  3 . Sensory assessments took place between 10–12 a.m. The beers were served in black-colored glasses. Per session, between 5 and 12 beers of the same style were tasted at 12 °C to 16 °C. Two reference beers were added to each set and indicated as ‘Reference 1 & 2’, allowing panel members to calibrate their ratings. Not all panelists were present at every tasting. Scores were scaled by standard deviation and mean-centered per taster. Values are represented as z-scores and clustered by Euclidean distance. Pairwise Spearman correlations were calculated between taste and aroma sensory attributes. Panel consistency was evaluated by repeating samples on different sessions and performing ANOVA to identify differences, using the ‘stats’ package (v4.2.2) in R (for package information, see Supplementary Table  S8 ).

Online reviews from a public database

The ‘scrapy’ package in Python (v3.6) (for package information, see Supplementary Table  S8 ). was used to collect 232,288 online reviews (mean=922, min=6, max=5343) from RateBeer, an online beer review database. Each review entry comprised 5 numerical scores (appearance, aroma, taste, palate and overall quality) and an optional review text. The total number of reviews per reviewer was collected separately. Numerical scores were scaled and centered per rater, and mean scores were calculated per beer.

For the review texts, the language was estimated using the packages ‘langdetect’ and ‘langid’ in Python. Reviews that were classified as English by both packages were kept. Reviewers with fewer than 100 entries overall were discarded. 181,025 reviews from >6000 reviewers from >40 countries remained. Text processing was done using the ‘nltk’ package in Python. Texts were corrected for slang and misspellings; proper nouns and rare words that are relevant to the beer context were specified and kept as-is (‘Chimay’,’Lambic’, etc.). A dictionary of semantically similar sensorial terms, for example ‘floral’ and ‘flower’, was created and collapsed together into one term. Words were stemmed and lemmatized to avoid identifying words such as ‘acid’ and ‘acidity’ as separate terms. Numbers and punctuation were removed.

Sentences from up to 50 randomly chosen reviews per beer were manually categorized according to the aspect of beer they describe (appearance, aroma, taste, palate, overall quality—not to be confused with the 5 numerical scores described above) or flagged as irrelevant if they contained no useful information. If a beer contained fewer than 50 reviews, all reviews were manually classified. This labeled data set was used to train a model that classified the rest of the sentences for all beers 91 . Sentences describing taste and aroma were extracted, and term frequency–inverse document frequency (TFIDF) was implemented to calculate enrichment scores for sensorial words per beer.

The sex of the tasting subject was not considered when building our sensory database. Instead, results from different panelists were averaged, both for our trained panel (56% male, 44% female) and the RateBeer reviews (70% male, 30% female for RateBeer as a whole).

Beer price collection and processing

Beer prices were collected from the following stores: Colruyt, Delhaize, Total Wine, BeerHawk, The Belgian Beer Shop, The Belgian Shop, and Beer of Belgium. Where applicable, prices were converted to Euros and normalized per liter. Spearman correlations were calculated between these prices and mean overall appreciation scores from RateBeer and the taste panel, respectively.

Pairwise Spearman Rank correlations were calculated between all sensory properties.

Machine learning models

Predictive modeling of sensory profiles from chemical data.

Regression models were constructed to predict (a) trained panel scores for beer flavors and quality from beer chemical profiles and (b) public reviews’ appreciation scores from beer chemical profiles. Z-scores were used to represent sensory attributes in both data sets. Chemical properties with log-normal distributions (Shapiro-Wilk test, p  <  0.05 ) were log-transformed. Missing chemical measurements (0.1% of all data) were replaced with mean values per attribute. Observations from 250 beers were randomly separated into a training set (70%, 175 beers) and a test set (30%, 75 beers), stratified per beer style. Chemical measurements (p = 231) were normalized based on the training set average and standard deviation. In total, three linear regression-based models: linear regression with first-order interaction terms (LR), lasso regression with first-order interaction terms (Lasso) and partial least squares regression (PLSR); five decision tree models, Adaboost regressor (ABR), Extra Trees (ET), Gradient Boosting regressor (GBR), Random Forest (RF) and XGBoost regressor (XGBR); one support vector machine model (SVR) and one artificial neural network model (ANN) were trained. The models were implemented using the ‘scikit-learn’ package (v1.2.2) and ‘xgboost’ package (v1.7.3) in Python (v3.9.16). Models were trained, and hyperparameters optimized, using five-fold cross-validated grid search with the coefficient of determination (R 2 ) as the evaluation metric. The ANN (scikit-learn’s MLPRegressor) was optimized using Bayesian Tree-Structured Parzen Estimator optimization with the ‘Optuna’ Python package (v3.2.0). Individual models were trained per attribute, and a multi-output model was trained on all attributes simultaneously.

Model dissection

GBR was found to outperform other methods, resulting in models with the highest average R 2 values in both trained panel and public review data sets. Impurity-based rankings of the most important predictors for each predicted sensorial trait were obtained using the ‘scikit-learn’ package. To observe the relationships between these chemical properties and their predicted targets, partial dependence plots (PDP) were constructed for the six most important predictors of consumer appreciation 74 , 75 .

The ‘SHAP’ package in Python (v0.41.0) was implemented to provide an alternative ranking of predictor importance and to visualize the predictors’ effects as a function of their concentration 68 .

Validation of causal chemical properties

To validate the effects of the most important model features on predicted sensory attributes, beers were spiked with the chemical compounds identified by the models and descriptive sensory analyses were carried out according to the American Society of Brewing Chemists (ASBC) protocol 90 .

Compound spiking was done 30 min before tasting. Compounds were spiked into fresh beer bottles, that were immediately resealed and inverted three times. Fresh bottles of beer were opened for the same duration, resealed, and inverted thrice, to serve as controls. Pairs of spiked samples and controls were served simultaneously, chilled and in dark glasses as outlined in the Trained panel section above. Tasters were instructed to select the glass with the higher flavor intensity for each attribute (directional difference test 92 ) and to select the glass they prefer.

The final concentration after spiking was equal to the within-style average, after normalizing by ethanol concentration. This was done to ensure balanced flavor profiles in the final spiked beer. The same methods were applied to improve a non-alcoholic beer. Compounds were the following: ethyl acetate (Merck KGaA, W241415), ethyl hexanoate (Merck KGaA, W243906), isoamyl acetate (Merck KGaA, W205508), phenethyl acetate (Merck KGaA, W285706), ethanol (96%, Colruyt), glycerol (Merck KGaA, W252506), lactic acid (Merck KGaA, 261106).

Significant differences in preference or perceived intensity were determined by performing the two-sided binomial test on each attribute.

Reporting summary

Further information on research design is available in the  Nature Portfolio Reporting Summary linked to this article.

Data availability

The data that support the findings of this work are available in the Supplementary Data files and have been deposited to Zenodo under accession code 10653704 93 . The RateBeer scores data are under restricted access, they are not publicly available as they are property of RateBeer (ZX Ventures, USA). Access can be obtained from the authors upon reasonable request and with permission of RateBeer (ZX Ventures, USA).  Source data are provided with this paper.

Code availability

The code for training the machine learning models, analyzing the models, and generating the figures has been deposited to Zenodo under accession code 10653704 93 .

Tieman, D. et al. A chemical genetic roadmap to improved tomato flavor. Science 355 , 391–394 (2017).

Article   ADS   CAS   PubMed   Google Scholar  

Plutowska, B. & Wardencki, W. Application of gas chromatography–olfactometry (GC–O) in analysis and quality assessment of alcoholic beverages – A review. Food Chem. 107 , 449–463 (2008).

Article   CAS   Google Scholar  

Legin, A., Rudnitskaya, A., Seleznev, B. & Vlasov, Y. Electronic tongue for quality assessment of ethanol, vodka and eau-de-vie. Anal. Chim. Acta 534 , 129–135 (2005).

Loutfi, A., Coradeschi, S., Mani, G. K., Shankar, P. & Rayappan, J. B. B. Electronic noses for food quality: A review. J. Food Eng. 144 , 103–111 (2015).

Ahn, Y.-Y., Ahnert, S. E., Bagrow, J. P. & Barabási, A.-L. Flavor network and the principles of food pairing. Sci. Rep. 1 , 196 (2011).

Article   CAS   PubMed   PubMed Central   Google Scholar  

Bartoshuk, L. M. & Klee, H. J. Better fruits and vegetables through sensory analysis. Curr. Biol. 23 , R374–R378 (2013).

Article   CAS   PubMed   Google Scholar  

Piggott, J. R. Design questions in sensory and consumer science. Food Qual. Prefer. 3293 , 217–220 (1995).

Article   Google Scholar  

Kermit, M. & Lengard, V. Assessing the performance of a sensory panel-panellist monitoring and tracking. J. Chemom. 19 , 154–161 (2005).

Cook, D. J., Hollowood, T. A., Linforth, R. S. T. & Taylor, A. J. Correlating instrumental measurements of texture and flavour release with human perception. Int. J. Food Sci. Technol. 40 , 631–641 (2005).

Chinchanachokchai, S., Thontirawong, P. & Chinchanachokchai, P. A tale of two recommender systems: The moderating role of consumer expertise on artificial intelligence based product recommendations. J. Retail. Consum. Serv. 61 , 1–12 (2021).

Ross, C. F. Sensory science at the human-machine interface. Trends Food Sci. Technol. 20 , 63–72 (2009).

Chambers, E. IV & Koppel, K. Associations of volatile compounds with sensory aroma and flavor: The complex nature of flavor. Molecules 18 , 4887–4905 (2013).

Pinu, F. R. Metabolomics—The new frontier in food safety and quality research. Food Res. Int. 72 , 80–81 (2015).

Danezis, G. P., Tsagkaris, A. S., Brusic, V. & Georgiou, C. A. Food authentication: state of the art and prospects. Curr. Opin. Food Sci. 10 , 22–31 (2016).

Shepherd, G. M. Smell images and the flavour system in the human brain. Nature 444 , 316–321 (2006).

Meilgaard, M. C. Prediction of flavor differences between beers from their chemical composition. J. Agric. Food Chem. 30 , 1009–1017 (1982).

Xu, L. et al. Widespread receptor-driven modulation in peripheral olfactory coding. Science 368 , eaaz5390 (2020).

Kupferschmidt, K. Following the flavor. Science 340 , 808–809 (2013).

Billesbølle, C. B. et al. Structural basis of odorant recognition by a human odorant receptor. Nature 615 , 742–749 (2023).

Article   ADS   PubMed   PubMed Central   Google Scholar  

Smith, B. Perspective: Complexities of flavour. Nature 486 , S6–S6 (2012).

Pfister, P. et al. Odorant receptor inhibition is fundamental to odor encoding. Curr. Biol. 30 , 2574–2587 (2020).

Moskowitz, H. W., Kumaraiah, V., Sharma, K. N., Jacobs, H. L. & Sharma, S. D. Cross-cultural differences in simple taste preferences. Science 190 , 1217–1218 (1975).

Eriksson, N. et al. A genetic variant near olfactory receptor genes influences cilantro preference. Flavour 1 , 22 (2012).

Ferdenzi, C. et al. Variability of affective responses to odors: Culture, gender, and olfactory knowledge. Chem. Senses 38 , 175–186 (2013).

Article   PubMed   Google Scholar  

Lawless, H. T. & Heymann, H. Sensory evaluation of food: Principles and practices. (Springer, New York, NY). https://doi.org/10.1007/978-1-4419-6488-5 (2010).

Colantonio, V. et al. Metabolomic selection for enhanced fruit flavor. Proc. Natl. Acad. Sci. 119 , e2115865119 (2022).

Fritz, F., Preissner, R. & Banerjee, P. VirtualTaste: a web server for the prediction of organoleptic properties of chemical compounds. Nucleic Acids Res 49 , W679–W684 (2021).

Tuwani, R., Wadhwa, S. & Bagler, G. BitterSweet: Building machine learning models for predicting the bitter and sweet taste of small molecules. Sci. Rep. 9 , 1–13 (2019).

Dagan-Wiener, A. et al. Bitter or not? BitterPredict, a tool for predicting taste from chemical structure. Sci. Rep. 7 , 1–13 (2017).

Pallante, L. et al. Toward a general and interpretable umami taste predictor using a multi-objective machine learning approach. Sci. Rep. 12 , 1–11 (2022).

Malavolta, M. et al. A survey on computational taste predictors. Eur. Food Res. Technol. 248 , 2215–2235 (2022).

Lee, B. K. et al. A principal odor map unifies diverse tasks in olfactory perception. Science 381 , 999–1006 (2023).

Mayhew, E. J. et al. Transport features predict if a molecule is odorous. Proc. Natl. Acad. Sci. 119 , e2116576119 (2022).

Niu, Y. et al. Sensory evaluation of the synergism among ester odorants in light aroma-type liquor by odor threshold, aroma intensity and flash GC electronic nose. Food Res. Int. 113 , 102–114 (2018).

Yu, P., Low, M. Y. & Zhou, W. Design of experiments and regression modelling in food flavour and sensory analysis: A review. Trends Food Sci. Technol. 71 , 202–215 (2018).

Oladokun, O. et al. The impact of hop bitter acid and polyphenol profiles on the perceived bitterness of beer. Food Chem. 205 , 212–220 (2016).

Linforth, R., Cabannes, M., Hewson, L., Yang, N. & Taylor, A. Effect of fat content on flavor delivery during consumption: An in vivo model. J. Agric. Food Chem. 58 , 6905–6911 (2010).

Guo, S., Na Jom, K. & Ge, Y. Influence of roasting condition on flavor profile of sunflower seeds: A flavoromics approach. Sci. Rep. 9 , 11295 (2019).

Ren, Q. et al. The changes of microbial community and flavor compound in the fermentation process of Chinese rice wine using Fagopyrum tataricum grain as feedstock. Sci. Rep. 9 , 3365 (2019).

Hastie, T., Friedman, J. & Tibshirani, R. The Elements of Statistical Learning. (Springer, New York, NY). https://doi.org/10.1007/978-0-387-21606-5 (2001).

Dietz, C., Cook, D., Huismann, M., Wilson, C. & Ford, R. The multisensory perception of hop essential oil: a review. J. Inst. Brew. 126 , 320–342 (2020).

CAS   Google Scholar  

Roncoroni, Miguel & Verstrepen, Kevin Joan. Belgian Beer: Tested and Tasted. (Lannoo, 2018).

Meilgaard, M. Flavor chemistry of beer: Part II: Flavor and threshold of 239 aroma volatiles. in (1975).

Bokulich, N. A. & Bamforth, C. W. The microbiology of malting and brewing. Microbiol. Mol. Biol. Rev. MMBR 77 , 157–172 (2013).

Dzialo, M. C., Park, R., Steensels, J., Lievens, B. & Verstrepen, K. J. Physiology, ecology and industrial applications of aroma formation in yeast. FEMS Microbiol. Rev. 41 , S95–S128 (2017).

Article   PubMed   PubMed Central   Google Scholar  

Datta, A. et al. Computer-aided food engineering. Nat. Food 3 , 894–904 (2022).

American Society of Brewing Chemists. Beer Methods. (American Society of Brewing Chemists, St. Paul, MN, U.S.A.).

Olaniran, A. O., Hiralal, L., Mokoena, M. P. & Pillay, B. Flavour-active volatile compounds in beer: production, regulation and control. J. Inst. Brew. 123 , 13–23 (2017).

Verstrepen, K. J. et al. Flavor-active esters: Adding fruitiness to beer. J. Biosci. Bioeng. 96 , 110–118 (2003).

Meilgaard, M. C. Flavour chemistry of beer. part I: flavour interaction between principal volatiles. Master Brew. Assoc. Am. Tech. Q 12 , 107–117 (1975).

Briggs, D. E., Boulton, C. A., Brookes, P. A. & Stevens, R. Brewing 227–254. (Woodhead Publishing). https://doi.org/10.1533/9781855739062.227 (2004).

Bossaert, S., Crauwels, S., De Rouck, G. & Lievens, B. The power of sour - A review: Old traditions, new opportunities. BrewingScience 72 , 78–88 (2019).

Google Scholar  

Verstrepen, K. J. et al. Flavor active esters: Adding fruitiness to beer. J. Biosci. Bioeng. 96 , 110–118 (2003).

Snauwaert, I. et al. Microbial diversity and metabolite composition of Belgian red-brown acidic ales. Int. J. Food Microbiol. 221 , 1–11 (2016).

Spitaels, F. et al. The microbial diversity of traditional spontaneously fermented lambic beer. PLoS ONE 9 , e95384 (2014).

Blanco, C. A., Andrés-Iglesias, C. & Montero, O. Low-alcohol Beers: Flavor Compounds, Defects, and Improvement Strategies. Crit. Rev. Food Sci. Nutr. 56 , 1379–1388 (2016).

Jackowski, M. & Trusek, A. Non-Alcohol. beer Prod. – Overv. 20 , 32–38 (2018).

Takoi, K. et al. The contribution of geraniol metabolism to the citrus flavour of beer: Synergy of geraniol and β-citronellol under coexistence with excess linalool. J. Inst. Brew. 116 , 251–260 (2010).

Kroeze, J. H. & Bartoshuk, L. M. Bitterness suppression as revealed by split-tongue taste stimulation in humans. Physiol. Behav. 35 , 779–783 (1985).

Mennella, J. A. et al. A spoonful of sugar helps the medicine go down”: Bitter masking bysucrose among children and adults. Chem. Senses 40 , 17–25 (2015).

Wietstock, P., Kunz, T., Perreira, F. & Methner, F.-J. Metal chelation behavior of hop acids in buffered model systems. BrewingScience 69 , 56–63 (2016).

Sancho, D., Blanco, C. A., Caballero, I. & Pascual, A. Free iron in pale, dark and alcohol-free commercial lager beers. J. Sci. Food Agric. 91 , 1142–1147 (2011).

Rodrigues, H. & Parr, W. V. Contribution of cross-cultural studies to understanding wine appreciation: A review. Food Res. Int. 115 , 251–258 (2019).

Korneva, E. & Blockeel, H. Towards better evaluation of multi-target regression models. in ECML PKDD 2020 Workshops (eds. Koprinska, I. et al.) 353–362 (Springer International Publishing, Cham, 2020). https://doi.org/10.1007/978-3-030-65965-3_23 .

Gastón Ares. Mathematical and Statistical Methods in Food Science and Technology. (Wiley, 2013).

Grinsztajn, L., Oyallon, E. & Varoquaux, G. Why do tree-based models still outperform deep learning on tabular data? Preprint at http://arxiv.org/abs/2207.08815 (2022).

Gries, S. T. Statistics for Linguistics with R: A Practical Introduction. in Statistics for Linguistics with R (De Gruyter Mouton, 2021). https://doi.org/10.1515/9783110718256 .

Lundberg, S. M. et al. From local explanations to global understanding with explainable AI for trees. Nat. Mach. Intell. 2 , 56–67 (2020).

Ickes, C. M. & Cadwallader, K. R. Effects of ethanol on flavor perception in alcoholic beverages. Chemosens. Percept. 10 , 119–134 (2017).

Kato, M. et al. Influence of high molecular weight polypeptides on the mouthfeel of commercial beer. J. Inst. Brew. 127 , 27–40 (2021).

Wauters, R. et al. Novel Saccharomyces cerevisiae variants slow down the accumulation of staling aldehydes and improve beer shelf-life. Food Chem. 398 , 1–11 (2023).

Li, H., Jia, S. & Zhang, W. Rapid determination of low-level sulfur compounds in beer by headspace gas chromatography with a pulsed flame photometric detector. J. Am. Soc. Brew. Chem. 66 , 188–191 (2008).

Dercksen, A., Laurens, J., Torline, P., Axcell, B. C. & Rohwer, E. Quantitative analysis of volatile sulfur compounds in beer using a membrane extraction interface. J. Am. Soc. Brew. Chem. 54 , 228–233 (1996).

Molnar, C. Interpretable Machine Learning: A Guide for Making Black-Box Models Interpretable. (2020).

Zhao, Q. & Hastie, T. Causal interpretations of black-box models. J. Bus. Econ. Stat. Publ. Am. Stat. Assoc. 39 , 272–281 (2019).

Article   MathSciNet   Google Scholar  

Hastie, T., Tibshirani, R. & Friedman, J. The Elements of Statistical Learning. (Springer, 2019).

Labrado, D. et al. Identification by NMR of key compounds present in beer distillates and residual phases after dealcoholization by vacuum distillation. J. Sci. Food Agric. 100 , 3971–3978 (2020).

Lusk, L. T., Kay, S. B., Porubcan, A. & Ryder, D. S. Key olfactory cues for beer oxidation. J. Am. Soc. Brew. Chem. 70 , 257–261 (2012).

Gonzalez Viejo, C., Torrico, D. D., Dunshea, F. R. & Fuentes, S. Development of artificial neural network models to assess beer acceptability based on sensory properties using a robotic pourer: A comparative model approach to achieve an artificial intelligence system. Beverages 5 , 33 (2019).

Gonzalez Viejo, C., Fuentes, S., Torrico, D. D., Godbole, A. & Dunshea, F. R. Chemical characterization of aromas in beer and their effect on consumers liking. Food Chem. 293 , 479–485 (2019).

Gilbert, J. L. et al. Identifying breeding priorities for blueberry flavor using biochemical, sensory, and genotype by environment analyses. PLOS ONE 10 , 1–21 (2015).

Goulet, C. et al. Role of an esterase in flavor volatile variation within the tomato clade. Proc. Natl. Acad. Sci. 109 , 19009–19014 (2012).

Article   ADS   CAS   PubMed   PubMed Central   Google Scholar  

Borisov, V. et al. Deep Neural Networks and Tabular Data: A Survey. IEEE Trans. Neural Netw. Learn. Syst. 1–21 https://doi.org/10.1109/TNNLS.2022.3229161 (2022).

Statista. Statista Consumer Market Outlook: Beer - Worldwide.

Seitz, H. K. & Stickel, F. Molecular mechanisms of alcoholmediated carcinogenesis. Nat. Rev. Cancer 7 , 599–612 (2007).

Voordeckers, K. et al. Ethanol exposure increases mutation rate through error-prone polymerases. Nat. Commun. 11 , 3664 (2020).

Goelen, T. et al. Bacterial phylogeny predicts volatile organic compound composition and olfactory response of an aphid parasitoid. Oikos 129 , 1415–1428 (2020).

Article   ADS   Google Scholar  

Reher, T. et al. Evaluation of hop (Humulus lupulus) as a repellent for the management of Drosophila suzukii. Crop Prot. 124 , 104839 (2019).

Stein, S. E. An integrated method for spectrum extraction and compound identification from gas chromatography/mass spectrometry data. J. Am. Soc. Mass Spectrom. 10 , 770–781 (1999).

American Society of Brewing Chemists. Sensory Analysis Methods. (American Society of Brewing Chemists, St. Paul, MN, U.S.A., 1992).

McAuley, J., Leskovec, J. & Jurafsky, D. Learning Attitudes and Attributes from Multi-Aspect Reviews. Preprint at https://doi.org/10.48550/arXiv.1210.3926 (2012).

Meilgaard, M. C., Carr, B. T. & Carr, B. T. Sensory Evaluation Techniques. (CRC Press, Boca Raton). https://doi.org/10.1201/b16452 (2014).

Schreurs, M. et al. Data from: Predicting and improving complex beer flavor through machine learning. Zenodo https://doi.org/10.5281/zenodo.10653704 (2024).

Download references

Acknowledgements

We thank all lab members for their discussions and thank all tasting panel members for their contributions. Special thanks go out to Dr. Karin Voordeckers for her tremendous help in proofreading and improving the manuscript. M.S. was supported by a Baillet-Latour fellowship, L.C. acknowledges financial support from KU Leuven (C16/17/006), F.A.T. was supported by a PhD fellowship from FWO (1S08821N). Research in the lab of K.J.V. is supported by KU Leuven, FWO, VIB, VLAIO and the Brewing Science Serves Health Fund. Research in the lab of T.W. is supported by FWO (G.0A51.15) and KU Leuven (C16/17/006).

Author information

These authors contributed equally: Michiel Schreurs, Supinya Piampongsant, Miguel Roncoroni.

Authors and Affiliations

VIB—KU Leuven Center for Microbiology, Gaston Geenslaan 1, B-3001, Leuven, Belgium

Michiel Schreurs, Supinya Piampongsant, Miguel Roncoroni, Lloyd Cool, Beatriz Herrera-Malaver, Florian A. Theßeling & Kevin J. Verstrepen

CMPG Laboratory of Genetics and Genomics, KU Leuven, Gaston Geenslaan 1, B-3001, Leuven, Belgium

Leuven Institute for Beer Research (LIBR), Gaston Geenslaan 1, B-3001, Leuven, Belgium

Laboratory of Socioecology and Social Evolution, KU Leuven, Naamsestraat 59, B-3000, Leuven, Belgium

Lloyd Cool, Christophe Vanderaa & Tom Wenseleers

VIB Bioinformatics Core, VIB, Rijvisschestraat 120, B-9052, Ghent, Belgium

Łukasz Kreft & Alexander Botzki

AB InBev SA/NV, Brouwerijplein 1, B-3000, Leuven, Belgium

Philippe Malcorps & Luk Daenen

You can also search for this author in PubMed   Google Scholar

Contributions

S.P., M.S. and K.J.V. conceived the experiments. S.P., M.S. and K.J.V. designed the experiments. S.P., M.S., M.R., B.H. and F.A.T. performed the experiments. S.P., M.S., L.C., C.V., L.K., A.B., P.M., L.D., T.W. and K.J.V. contributed analysis ideas. S.P., M.S., L.C., C.V., T.W. and K.J.V. analyzed the data. All authors contributed to writing the manuscript.

Corresponding author

Correspondence to Kevin J. Verstrepen .

Ethics declarations

Competing interests.

K.J.V. is affiliated with bar.on. The other authors declare no competing interests.

Peer review

Peer review information.

Nature Communications thanks Florian Bauer, Andrew John Macintosh and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary information, peer review file, description of additional supplementary files, supplementary data 1, supplementary data 2, supplementary data 3, supplementary data 4, supplementary data 5, supplementary data 6, supplementary data 7, reporting summary, source data, source data, rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Schreurs, M., Piampongsant, S., Roncoroni, M. et al. Predicting and improving complex beer flavor through machine learning. Nat Commun 15 , 2368 (2024). https://doi.org/10.1038/s41467-024-46346-0

Download citation

Received : 30 October 2023

Accepted : 21 February 2024

Published : 26 March 2024

DOI : https://doi.org/10.1038/s41467-024-46346-0

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing: Translational Research newsletter — top stories in biotechnology, drug discovery and pharma.

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List

Artificial Intelligence in Nutrients Science Research: A Review

Jarosław sak.

1 Chair and Department of Humanities and Social Medicine, Medical University of Lublin, 20-093 Lublin, Poland

2 BioMolecular Resources Research Infrastructure Poland (BBMRI.pl), Poland

Magdalena Suchodolska

3 Faculty of Medicine, Medical University of Lublin, 20-059 Lublin, Poland; [email protected]

Artificial intelligence (AI) as a branch of computer science, the purpose of which is to imitate thought processes, learning abilities and knowledge management, finds more and more applications in experimental and clinical medicine. In recent decades, there has been an expansion of AI applications in biomedical sciences. The possibilities of artificial intelligence in the field of medical diagnostics, risk prediction and support of therapeutic techniques are growing rapidly. The aim of the article is to analyze the current use of AI in nutrients science research. The literature review was conducted in PubMed. A total of 399 records published between 1987 and 2020 were obtained, of which, after analyzing the titles and abstracts, 261 were rejected. In the next stages, the remaining records were analyzed using the full-text versions and, finally, 55 papers were selected. These papers were divided into three areas: AI in biomedical nutrients research (20 studies), AI in clinical nutrients research (22 studies) and AI in nutritional epidemiology (13 studies). It was found that the artificial neural network (ANN) methodology was dominant in the group of research on food composition study and production of nutrients. However, machine learning (ML) algorithms were widely used in studies on the influence of nutrients on the functioning of the human body in health and disease and in studies on the gut microbiota. Deep learning (DL) algorithms prevailed in a group of research works on clinical nutrients intake. The development of dietary systems using AI technology may lead to the creation of a global network that will be able to both actively support and monitor the personalized supply of nutrients.

1. Introduction

The term “artificial intelligence” was first proposed in 1955 by the American computer scientist John McCarthy (1927–2011) in the proposal of a research project, which was carried out the following year at Dartmouth College in Hanover, New Hampshire [ 1 , 2 ].

Artificial intelligence (AI) as a branch of computer science, the purpose of which is to imitate thought processes, learning abilities and knowledge management, finds more and more applications in experimental and clinical medicine. In recent decades, there has been an expansion of AI applications in medicine and biomedical sciences. The possibilities of artificial intelligence in the field of medical diagnostics, risk prediction and support of therapeutic techniques are growing rapidly. Thanks to the use of AI in ophthalmological [ 3 ], radiological [ 4 ] and cardiac [ 5 ] diagnostics, measurable clinical benefits have been obtained. AI was used in research on new pharmaceuticals [ 6 ]. The development of AI also provides new opportunities for research on nutrients and medical sensing technology [ 7 ].

1.1. Artificial Neural Networks (ANNs)

ANNs as a currently widely used modeling technique in the field of AI were inspired by the structure of natural neurons of the human brain. ANNs are mathematical models designed to process and calculate input signals through rows of processing elements, called artificial neurons, connected to each other by artificial synapses. There are three types of layers forming ANNs. The input layer captures the raw data and passes them to the hidden layer. In this second layer, the learning process takes place. The results of the analysis are collected in the output layer and the output data are created. A neural network may consist of hundreds of single units. An ANN is a parameterized system that has weights as adjustable parameters. Due to the need for estimation of these parameters, ANNs require large training sets. ANNs acquire knowledge by detecting patterns and relationships between data, i.e., through experience, not as a result of programming.

An ANN reveals its particular usefulness in the case of the need for modeling datasets with non-linear dependencies. In solving biomedical problems, raw data can be both literature and experimental data. In the last two decades, ANNs have been used, among others, to create an experimental decision algorithm model open to improvement, aimed at evaluating the results of biochemical tests confronted with both reference values and clinical data [ 8 ]. This technique was also used in evaluation of cell culture cross-contamination levels based on mass spectrometric fingerprints of intact mammalian cells [ 9 ]. The particular usefulness of ANNs has been proven in pharmaceutical analyses [ 10 ]. An interesting application of ANNs is the prediction of the relationship between the Mediterranean dietary pattern, clinical characteristics and cognitive functions [ 11 ]. The usefulness of ANNs has been proven in body composition analyses, which have clearly non-linear characteristics [ 12 ]. Using ANN modeling, significant benefits can be obtained in clinical dietetics.

It is worth noting that the fuzzy logic methodology (FLM) can be combined with neural networks. The idea of this area of AI is to strive for greater accuracy, dimensionality and simplification of the structure. There is a possibility to create fuzzy neural networks and convert FLM-based models into neural networks.

1.2. Machine Learning (ML)

ML is an AI area related to algorithms that improve automatically through experience. ML algorithms have the potential to create mathematical models for decision making. The process of creating these models is based on large sets of training data, without programming. The popularization of the use of ML algorithms took place in the last decade of the 20th century in search engine applications. In the following decades, there were high hopes for significant discoveries in the field of organic synthesis with the use of increasingly advanced ML algorithms [ 13 ]. Despite the fact that these hopes have not been fully met, this area of AI has important applications both in biomedical sciences and in clinical medicine. Machine learning—both supervised and unsupervised—can be applied to clinical datasets to develop risk models [ 14 ]. It can significantly support the analysis of data obtained from the patient [ 15 ].

There are suggestions that ML is the future of computer-assisted diagnostics, biomedical research and personalized medicine [ 16 ]. Machine learning techniques are becoming more and more popular in diabetes research: in blood glucose prediction and in the development of the so-called artificial pancreas (a closed-loop system) [ 17 ]. The use of ML algorithms in research on the gut microbiota is postulated, especially because of the large datasets collected in these studies [ 18 ]. In a recent report, Liu et al. proved that an ML algorithm integrating baseline microbial signatures of the intestinal microbiota can accurately predict the patient’s glycemic response to physical effort [ 19 ].

Deep learning (DL) is a subtype of ML. It is an AI domain that has found its applications especially in the techniques of image and voice recognition and foreign language translation. DL also has an important use in medical diagnostics. The significant advantage of DL over supervised ML is expressed in the autonomy of the program in the area of building sets of features used in recognition.

1.3. Internet of Things (IoT)

The term IoT was first used by British entrepreneur and startup founder Kevin Ashton in 1999, in the sense of a network of connected objects. This is the concept that objects (devices) can directly or indirectly collect, process or exchange data via a computer network or intelligent electrical installation. The term Internet of Everything (IoE) is used to describe a network of people, processes, data and things connected to the Internet.

In clinical medicine, IoT has a significant application in relation to telemedicine procedures [ 20 , 21 ], which are becoming more and more widely used, especially during the COVID-19 pandemic. Important applications of IoT can also be seen in the provision of detailed information on food products available on the market [ 22 ].

2. Materials and Methods

The aim of the article is to analyze the current use of AI in nutrients science research and to determine the prospects of its further application in this area.

The literature review was conducted in PubMed using a combination of searching terms: “artificial intelligence” (All Fields) AND “nutrients” (All Fields). A total of 399 records (published between 1987 and 2020) were obtained, of which, after analyzing the titles and abstracts, 261 were rejected. In the next stage, the remaining records were analyzed using the full-text versions and 111 papers were selected. These papers were afterwards divided into four categories: AI in agricultural nutrients research, AI in biomedical, AI in clinical nutrients research and AI in nutritional epidemiology. In order to limit the analyzed issues to biomedical aspects, agricultural and environmental nutrients research was excluded ( Figure 1 ).

Methodological flowchart of papers reviewed.

3.1. AI in Biomedical Nutrients Research

In the area of biomedical nutrients research, there were identified studies in which advanced AI methods and systems were applied in relation to the study of the composition of food products, optimization of nutrient production, the effects of nutrients on the functioning of the human body in health and disease and research on the gut microbiota ( Table 1 ).

The characteristics of the included studies on biomedical nutrients research.

Note: Domains: ANN = artificial neural network, ML = machine learning, FLM = fuzzy logic methodology, NV = network visualization; learning algorithms: kNN = k-nearest neighbor, KohNN = Kohonen neural network, LM = Levenberg–Marquardt algorithm, GA = genetic algorithm, ANN-GAR = Garson’s algorithm, GA-Fuzzy = fuzzy genetic algorithm, FFD = fractional factorial design, LASSO = least absolute shrinkage and selection operator, GA-PLS = genetic algorithm-partial least squares, PLS = partial least squares regression, GA-RBFN = genetic algorithm-radial basis function network, LS-SVM = least squares support vector machine, SVM = support vector machine, SVR = support vector regression, BN = Bayes net, NB = naive Bayes, RF = random forest, CLAs = clustering algorithms.

According to graphical characteristics of the analyzed works ( Figure 2 ), the ANN methodology dominated both in food composition study and the production of nutrients. Among the works on the influence of nutrients on the functioning of the human body in health and disease and studies on the gut microbiota, ML domain algorithms were used almost exclusively. The fuzzy logic methodology was used occasionally.

The studies of nutrients science research in relation to artificial intelligence (AI) domains. Note: AI domains: ML = machine learning, DL = deep learning, ANN = artificial neural network, FLM = fuzzy logic methodology, IoT = Internet of Things; biomedical nutrients research: FC = food composition; PoN = production of nutrients; IoNoF= influence of nutrients on phys./path. functions; GtM= gut microbiota; clinical nutrients research: CNI = clinical nutrients intake; DRFN= diseases risks to food and nutrients patterns; DTE = disease and trace elements levels; SUPPL = supplementations; nutritional epidemiology: DAST = dietary assessment; PMSs = physical monitoring systems; ETEMS = environmental trace elements monitoring systems, […] = references.

3.1.1. AI in Food Composition Study

The use of AI techniques in studying the composition of food products and testing their originality dates back to the 1990s. Dettmar et al. used the ANN technique to identify the region of origin of fruit from a set of 16 variables characterizing samples of orange juice [ 23 ]. The effectiveness of the applied calculation technique was 92.5%.

Yang et al. used the isobaric tag for a relative and absolute quantification proteomic approach to analyze differentially expressed whey proteins in the human and bovine colostrum and mature milk to understand the different whey proteomes. It may provide useful information for the development of nutrient food for infants and dairy products [ 24 ].

Moreira et al. used topological maps of the Kohonen neural network in the assessment of the procedure for sample preparation of cashew nuts [ 25 ]. Shen et al. used laser-induced breakdown spectroscopy (LIBS), least squares support vector machines (LS-SVM) and LASSO models for the detection of six nutritive elements in Panax notoginseng (traditional Chinese medicine) samples from eight producing areas [ 26 ]. Rasouli et al. applied the whole space genetic algorithm-radial basis function network (wsGA-RBFN) method to determine the content of microminerals of Fe 2+ , Zn 2+ , Co 2+ and Cu 2+ in various pharmaceutical products and vegetable samples (tomato, lettuce, white and red cabbages) [ 27 ]. This group of studies also includes the research of Soltani et al. who used three different quantitative structure bitter taste relationship (QSBR) models (artificial neural network, multiple linear regression and support vector machine) to predict the bitterness of 229 peptides [ 28 ].

3.1.2. AI in Research on Production of Nutrients

With regard to research on the optimization of the production of certain nutrients, several studies have been identified in which AI modeling was intentionally applied.

Huang et al. implemented methods of production of a retinol derivative named retinyl laurate by an artificial neural network (ANN) [ 29 ]. Zheng et al. studied the optimization of producing 2,6-dimethoxy-ρ-benzoquinone (DMBQ) and methoxy-ρ-benzoquinone (MBQ) as the potential anticancer compounds in fermented wheat germ. They used algorithms of an artificial neural network (ANN) combined with the genetic algorithm (GA) [ 30 ]. The ANN model with a Levenberg–Marquardt training algorithm was applied for modeling the complicated non-linear interactions among 16 nutrients in this production process. Kumar et al. used GA-Fuzzy—an evolutionary algorithm comprised of the genetic algorithm (GA) and the fuzzy logic methodology (FLM)—for the optimization of the production of phycobiliproteins (PBPs) from cyanobacteria [ 31 ].

3.1.3. AI in Research on the Influence of Nutrients on Physiological and Pathophysiological Functions

The most numerous group of works presenting applications of AI models in biomedical nutrients research is research on vitamins.

Pavani et al. used the neuro-fuzzy model to investigate the influence of alterations in vitamin K (K1, K2 and K3) on modulating the warfarin dose requirement [ 32 ]. An AI model was used to predict the warfarin dose, and higher vitamin K1 was observed in the CYP4F2 V433M polymorphism in this study.

The use of AI techniques in research on the influence of vitamin D on the functioning of the human body was described in articles published in 2019. Yu et al. compared the expression profiles of miRNAs, lncRNAs, mRNAs and circRNAs, between 1,25-(OH) 2 D 3 -treated endothelial progenitor cells (EPCs) and control cells. They used bioinformatics analyses to identify differentially expressed RNAs and constructed the competing endogenous RNA (ceRNA) networks with Cytoscape software [ 33 ]. Zhang et al. investigated the effect of 1,25-dihydroxy-vitamin D3 (1,25-(OH) 2 D 3 ) on primary chondrocytes cultured from patients with an osteoarthritis protein–protein interaction (PPI) by a PPI network [ 34 ]. They suggested that their study might provide a theoretical basis for the use of vitamin D in treating osteoarthritis.

Kolhe et al. tried to verify the hypothesis that vitamin C mediates proliferation and differentiation of bone marrow stromal cells through miRNA regulation [ 35 ]. They performed bioinformatics analyses to identify novel target genes and signaling pathways. Gene Ontology word clouds were generated using the online Wordle software.

Huang et al. investigated an influence of the active ingredients of licorice (root of Glycyrrhiza glabra ) for muscle fatigue by RNA-Seq and bioinformatics analysis. They used a machine learning model and a docking tool to predict active ingredients. They identified hispaglabridin B (HB) as a possible inhibitor of FoxO1 which was useful for preventing muscle wasting in chronic kidney disease [ 36 ].

Li et al. investigated the effects and mechanism of Ginkgo biloba L. on Alzheimer’s disease by using compound-target-disease and compound-group-target-pathway (CGTP) network models [ 37 ].

Panwar et al. developed in silico models for predicting vitamin-interacting residues in a protein from its primary structure. They used machine learning techniques such as various classifiers of SVM, RandomForest, BayesNet, NaiveBayes, NaiveBayesMultinomial and ComplementNaiveBayes and position-specific scoring matrix (PSSM) features of protein sequences to identify vitamin-interacting residues in a protein [ 38 ]. Yu et al. used a new predictor, the TargetVita web server, and datasets for predicting protein–vitamin binding residues using protein sequences [ 39 ].

3.1.4. AI in Research on Gut Microbiota

In recent years, results of research on nutrients and the gut microbiota using AI techniques have been published.

Devika and Raman used genome-scale metabolic models to differentiate between 36 important Bifidobacterial strains [ 40 ]. Shima et al. performed analyses concerning the gut microbiota, based on a combination of machine learning and network visualization [ 41 ]. Mohammed and Guda used AI in the research on enzymes produced by strains of gut bacteria [ 42 ]. They developed ECemble, an approach to identify enzymes and study the human gut metabolic pathways. ECemble uses an ensemble of machine learning methods to predict and identify the enzyme classes. They identified 48 pathways that have at least one bacteria-encoded enzyme and are involved in metabolizing nutrients.

3.2. AI in Clinical Nutrients Research

In the past studies in the field of clinical nutrients research, AI techniques have been used in projects aimed at creating tools supporting dietary activities and in supplementation, as well as in the diagnosis and prediction of the risk of chronic diseases ( Table 2 ).

The characteristics of the included studies on clinical nutrients research.

Note: Domains/methods: ANN = artificial neural network, ML = machine learning, DL = deep learning, FLM = fuzzy logic methodology; learning algorithms: kNN = k-nearest neighbor, LASSO = least absolute shrinkage and selection operator, FFNN = feed forward neural network, LR = linear regression, RF = random forest, DTA = decision tree algorithm, SVM = support vector machines, NB = naive Bayes, CLAs = clustering algorithms.

According to the graphical characteristics of the analyzed works ( Figure 2 ), the DL methodology dominated in the group of studies on clinical nutrients intake. A marginal use of the fuzzy logic methodology was noted—it appeared only in one study.

3.2.1. AI in Clinical Nutrients Intake

Among the identified studies on the application of AI in clinical practice, there is a need to distinguish those that aimed to develop systems that monitor, support and modulate the nutrition of chronically ill people. Lu et al. presented a novel system based on AI to accurately estimate nutrient intake, by simply processing RGB depth image pairs captured before and after meal consumption [ 43 ]. Oka et al. compared AI-supported nutrition therapy with a mobile application ( n = 50) versus human nutrition therapy ( n = 50) in a randomized controlled trial [ 44 ]. An interesting technological solution in the AI area was used by Vasiloglou et al. in relation to the clinical problem of controlling carbohydrate intake in patients with type 1 diabetes. These authors used GoCARB as a computer vision-based smartphone system in determining plated meals’ carbohydrate content. In this study, the estimation of carbohydrate content in 54 plated meals made by GoCARB was compared to the estimation made by six experienced dietitians. It was found that GoCARB estimated the carbohydrate content with the same accuracy as professional nutritionists ( p = 0.93) [ 45 ].

Chin et al. tested the Automated Self-Administered 24-Hour Dietary Assessment Tool (ASA24) on the example of lactose with regard to the Nutrition Data System for Research (NDSR) [ 46 ]. ASA24, also known as food diaries, is a web-based tool that enables multiple, automatically coded, self-administered 24-h diet recalls. NDSR is a dietary analysis software application widely used for the collection and coding of 24-h dietary recalls and the analysis of menus. Nine machine learning models have been developed based on the nutrients common to ASA24 and the NCC database. The results obtained in this study suggest that computational methods can successfully estimate an NCC-exclusive nutrient for foods reported in ASA24.

In order to monitor eating behaviors, a rapid automatic bite detection algorithm (RABID) that extracts and processes skeletal features from videos was constructed. Konstantinidis et al. used it to analyze the eating behaviors of n = 59 patients (three types of dishes, 45 meals), the results of which showed an agreement between algorithmic and human annotations (Cohen’s kappa κ = 0.894; F1-score: 0.948) [ 47 ].

Chi et al. proposed a knowledge-based system (KBS) for patients with chronic kidney disease using the Web Ontology Language (OWL) and the Semantic Web Rule Language (SWRL) [ 48 ]. In order to evaluate the designed system in recommending appropriate food serving amounts from different food groups, information was collected from n = 84 patients. It was found that the OWL-based KBS can achieve accurate problem solving and reasoning questions while maintaining the ability to share and extend the knowledge base.

AI techniques can also be useful in diagnosing mild dehydration. Posada-Quintero et al., using machine learning, investigated the possibility of detecting mild dehydration with autonomic responses to cognitive stress ( n = 17) [ 49 ]. Taking into account the autonomic control indicators based on electrodermal activity (EDA) and pulse rate variability (PRV) in the Stroop test, they obtained 91.2% overall accuracy of mild dehydration detection.

In the area of AI applications in the improvement of dietary solutions, two articles describing prototype solutions should be mentioned. Khan and Hoffmann proposed a menu construction using an incremental knowledge acquisition system (MIKAS) [ 50 ]. This system asks the expert to provide an explanation for each of their actions, in order to include the explanation in its knowledge base, so MIKAS could in the future automatically perform them.

Fuzzy arithmetic has been used to create “Nutri-Educ”—software for proper balancing of meals, according to the energy needs of the patient. Heuristic search algorithms are used to find a set of actions, acceptable from a nutritional point of view, that will transform the initial meal into a well-balanced one [ 51 ].

Baek et al. applied the hybrid clustering-based food recommendation method that uses chronic disease-based clustering and a nutrition knowledge base [ 52 ]. Food products are grouped using the k-means algorithm and food and nutrient data system. Based on the created clusters and data on food preferences, a knowledge base on diet and nutrition is generated.

Mezgec and Koroušić Seljak introduced a new “NutriNet” tool for food image recognition based on a deep convolutional neural network architecture [ 53 ]. It was tested on a collection of 225,953 images (512 × 512 pixels) of 520 different foods and beverages. This tool with an implemented training component is used in practice as a part of a mobile app for the dietary assessment of Parkinson’s disease patients.

3.2.2. AI in Evaluating Diseases Risks in Relations to Food and Nutrients Patterns

AI techniques also appear to be useful in estimating the risk of health problems based on the analysis of dietary or supplementation patterns. Panaretos et al. used the k-nearest neighbors algorithm and random forests decision tree to assess the 10-year cardiometabolic risk in relation to nutrient and food patterns, n = 3042 (2001–2002) [ 54 ]. The authors of the study, using factor analysis, identified factors from foods and nutrients, respectively, explaining 54 and 65% of the total variation in intake. ML techniques were found to be superior compared with linear regression in health score classification.

Berry et al. in n = 1002 twins and unrelated healthy adults groups (PREDICT 1 study) assessed the inter-individual variability of postprandial metabolic responses (triglyceride, glucose, insulin) as potential risk factors for cardiometabolic diseases [ 55 ]. On the basis of conducted cohort studies, they developed a machine learning model that predicted both glycemic (r = 0.77) and triglyceride (r = 0.47) responses to food intake.

Naushad et al. developed a breast cancer prediction model based on an artificial neural network (ANN) to investigate how micronutrients (foliate, B12) modulate susceptibility to breast cancer [ 56 ]. The developed ANN model explained 94.2% variability in breast cancer prediction.

This group of studies also includes the article by Shiao et al., who examined n = 106 participants in multi-ethnic colorectal cancer families in terms of prognostic factors of healthy eating (HEI index) [ 57 ]. Machine learning validation procedures were applied, including the ensemble method, generalized regression prediction, elastic net and leave-one-out cross-validation methods.

3.2.3. AI in Studying the Relationships between Disease and Trace Elements Levels

In a review of AI application reports, there were identified articles examining the levels of selected trace elements in biological samples collected from patients with type 2 diabetes. Tan et al. examined the usefulness of machine learning (Adaboost) in combination with trace element analysis of hair samples in diagnosing CVD in clinical practice ( n = 124) [ 58 ]. The same authors examined the levels of several elements, including trace elements: lithium, zinc, chromium, copper, iron, manganese, nickel and vanadium, in whole blood of type 2 diabetes patients ( n = 53), comparing them with analogous data obtained from healthy people ( n = 105) [ 59 ]. In order to construct the model, they used Fisher linear discriminate analysis (FLDA), a support vector machine (SVM) and a decision tree (DT) for data analysis. In 2014, the results of the relationships between several element levels in hair/urine and diabetes mellitus ( n = 211) were published using ensemble and single support vector machine (SVM) algorithms as the classification tools [ 60 ].

In addition to the use of AI techniques in the study of the relationship between the risk of diabetes and trace elements, the study of relationships between schizophrenia risk and serum levels of macro and trace elements should also be noted. Lin et al. for this purpose used samples taken from 114 schizophrenia patients and 114 healthy controls and supervised learning methods [ 61 ]. The levels of 39 macro and trace elements were examined and the best prediction accuracies were achieved by support vector machines.

3.2.4. AI in Studying on Supplementations

Li et al., in a recent report, described the performed bioinformatics analysis and computation assays using a network pharmacology method to evaluate the properties of vitamin A as an anti-SARS-CoV-2 regimen [ 62 ]. A similar research goal was achieved by the team of Chen et al., who, using network analysis, tested the potential of a novel combination of vitamin C, curcumin and glycyrrhizic acid (VCG Plus) against CoV infection [ 63 ]. Further, using network analysis, Fan et al. attempted to identify a molecular mechanism delaying the onset of psychotic symptoms in Alzheimer’s disease in association with the use of vitamin D [ 64 ].

3.3. AI in Nutritional Epidemiology

In the area of nutritional epidemiology research, there were identified studies in which advanced AI methods and systems were applied in relation to the dietary assessment, physical monitoring systems and environmental trace elements monitoring systems ( Table 3 ).

The characteristics of the included studies on nutritional epidemiology.

Note: Domains/methods: ANN = artificial neural network, ML = machine learning, DL = deep learning, FLM = fuzzy logic methodology, IoT = Internet of Things; learning algorithms: ICP = iterative closest point algorithm, CLAs = clustering algorithms, kNN = k-nearest neighbor, SVM = support vector machine, BDLN = Bayesian deep learning network, PNN = probabilistic neural network, KohNN = Kohonen neural network, PLS = partial least squares regression.

In this research area, the algorithms of ML and DL were used predominantly ( Figure 2 ). The methodology of ANN was used in environmental trace elements monitoring systems. The application of the IoT methodology was noted in the physical monitoring systems topic.

3.3.1. AI in Dietary Assessment

Mobile applications based on systems using AI are of significant importance in the field of nutritional prophylaxis ( Table 3 ). In 2008, Sun et al. proposed an electronic photographic approach and associated image processing algorithms to estimate food portion size [ 65 ]. Lu et al., in a recent publication, offered goFOOD TM as a dietary assessment system based on AI. It can estimate the calorie and macronutrient content of a meal, on the sole basis of food images captured by a smartphone [ 66 ].

Yang et al. proposed a new methodological approach in the field of nutritional epidemiology, Ontology for Nutritional Epidemiology (ONE) [ 67 ]. It is a resource to automate data integration, browsing and searching. ONE can be used to assess reporting completeness in nutritional epidemiology.

Lo et al. created an objective dietary assessment system based on a distinct neural network [ 68 ]. They used a depth image, the whole 3D point cloud map and iterative closest point (ICP) algorithms to improve the dietary behavior management.

Fang et al. estimated food energy based on images and the generative adversarial network (GAN) architecture ( n = 45) [ 69 ].

Ji et al. assessed the relative validity of an image-based dietary assessment app—Keenoa—and a 3-day food diary in a sample of healthy Canadian adults ( n = 102) [ 70 ]. The authors in this randomized controlled trial showed that Keenoa had better validity at the group level than the individual level and it can be used when focusing on the dietary intake of the general population.

Hsu et al. used the fuzzy decision model to develop a web-based support system that searches food composition databases and calculates dietary intake [ 71 ]. This research project was carried out due to the lack of integrated databases for Chinese menus and the need for a decision-making tool for dietitians in Taiwan.

3.3.2. AI in Physical Monitoring Systems

AI techniques have found their application not only in monitoring the quality and quantity of nutrients, but also in terms of the level of their expenditure. In the face of the obesity epidemic, these AI applications are very important. Monogaran et al. described the use of a monitoring system as an effective diagnosis tool of physical activities by a wearable smart-log patch with Internet of Things (IoT) sensors [ 72 ]. The data were analyzed using edge computing on a Bayesian deep learning network (EC-BDLN). Tragomalu et al. analyzed e-health applications for the management of cardiometabolic risk factors in children and adolescents [ 73 ]. Ramyaa et al. tried to phenotype women based on dietary macronutrients and physical activity using machine learning, support vector machine (SVM), neural network and k-nearest neighbors (kNN) algorithms [ 74 ].

3.3.3. AI in Environmental Trace Elements Monitoring Systems

Novic and Groselj used an ANN to create a methodology for food specifications associated with the origin of food. The methodology was tested on honey samples collected by the TRACE UE project [ 75 ]. The data were collected from various regions of Europe and analyzed for the content of trace elements.

Research on the content of trace elements and rare-earth elements in honey was also carried out by Drivelos et al. [ 76 ] They used probabilistic neural network (PNN) analysis and constructed a partial least squares (PLS) model for classifying of honey samples according to their geographical origin and organic characterization.

Tunakova et al. used an ANN to create a neural network model describing the retentions of trace elements in the human body. They calculated the microelement levels in the body, knowing the trace element levels in drinking water and urine [ 77 ].

4. Discussion

One of the main problems in analyzing publications on the use of AI in nutrient research is the range of research areas to be considered. This type of research creates a very diverse spectrum of problems. They are not limited to the field of biomedical sciences, but also apply to plant and animal breeding, including the breeding of microorganisms. The limitations which were found in the methodology of the review were dictated by the intention to maintain transparency. Therefore, studies that directly or indirectly relate to human health were included, excluding research on nutrients in agricultural and veterinary sciences. The review of the publications revealed three application areas of AI technology: biomedical nutrients research, clinical nutrients research and nutritional epidemiology.

During the analysis of the reviewed publications presenting the results of research on nutrients with the use of AI technology, it can be noticed a little later that it gained wider application in human health research than analogous applications in experimental research on food. This may have resulted from both some ethical concerns and psychological resistance, as well as from the imperfections of earlier AI algorithms, which seemed not yet ready to solve problems concerning the human body. A significant increase in the number of publications on the use of AI in nutrients research has been recorded in the last decade (2011–2020). Perhaps the title question from the article by Gedrich et al., “How optimal are computer-calculated optimal diets?” [ 78 ], asked at the end of the last century was significantly ahead of the medical professions’ mentality.

The use of AI in biomedical nutrients research reflects the need for efficient analysis of large datasets that could not be analyzed using traditional statistical methods. This applies in particular to the study of the relationship between nutrients and the functioning of the human body and in the study of the gut microbiota [ 40 , 41 , 42 ]. The increasing use of AI algorithms in this area is an expression of scientific progress and is becoming not only a privilege, but even a necessity in the pursuit of obtaining valuable results. The possible decoding of the gut microbiota functioning mechanisms can bring significant benefits in the form of possibilities to develop modern and very effective probiotics.

The application of AI algorithms in clinical nutrients research is expressed both by systems supporting dietary activities, diseases risks in relation to food and nutrients patterns and supplementation research. An important issue in this research area is the assessment of the reliability and credibility of the test results obtained using AI techniques. Another essential issue is the modification of the dietician–patient relationship in the case of replacing, in whole or in part, the work of a medical professional by AI systems [ 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 ]. The problem of trust in AI-based systems, especially in the elderly, remains open. In the social dimension, however, with the implementation of modern technologies in everyday activities, an increase in trust in both robotic systems and AI systems in medicine is observed. Especially on the basis of the articles included in the review, it is possible to state potentially good-quality effects of using dietary AI systems. Comparing them with the assessment of professional nutritionists, it is worth noting that in both cases, there were similar difficulties with regard to estimating the caloric value of some food products (e.g., GoCARB) [ 45 ]. The use of AI systems in dietary assessments enables personalized nutrition, which in some diseases is a priority.

The development of AI systems in dietetics may lead, in the near future, to a partial replacement of medical personnel and reducing the need for personal contact with a nutritionist. In the face of contemporary epidemiological threats, this seems to be of significant importance. The further dynamic development of dietary systems using AI technology may lead to the creation of a global network that will be able to both actively support and monitor the personalized supply of nutrients [ 79 ]. In this case, consideration should be given to geographical and cultural differences in the management of food and nutrients. Perhaps the development of AI in nutrients research will enable the creation of personalized nutrition databases as a starting point for modulating daily nutrition, as enabled by Nutri-Educ based on fuzzy arithmetic [ 51 ].

On the basis of this review, it is worthwhile to consider the possibility of creating AI systems to coordinate both biomedical and clinical nutrients research with nutritional epidemiology. Perhaps the gut microbiota function may be an important mediator of this kind of advanced coordination. Therefore, research on the importance of the intestinal flora is of fundamental importance in the field of nutrients research. A significant challenge for the near future is the use of AI technology in the creation of gut microbiota biobanks for the purpose of scientific research [ 80 ].

Despite the fact that AI technologies are dynamically developing, the problem in nutrients research is not currently obtaining more and more advanced algorithms, but the application of those that have already been developed and are standardly used in other fields of knowledge, and even in other areas of biomedicine. An important challenge for nutrients research is also their integration with research on the use of medical robotics. Perhaps the development and application of AI in nutrients research requires modification of both mentality and professional competences, as is already postulated in relation to the food industry [ 81 ].

Acknowledgments

This study was supported by grants from the Medical University of Lublin, no. DS 507.

Author Contributions

J.S. conceptualization, methodology, created the Figures and funding acquisition; J.S. and M.S. formal analysis, investigation, resources, analyzed the data, writing—original draft preparation, review and editing. All authors have read and agreed to the published version of the manuscript.

This research was funded by grants from the Medical University of Lublin, no. DS 507.

Conflicts of Interest

The authors declare no conflict of interest.

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

• Login To RMS System
• About JETIR URP
• About All Approval and Licence
• Conference/Special Issue Proposal
• Book and Dissertation/Thesis Publication
• How start New Journal & Software
• Best Papers Award
• Mission and Vision
• Reviewer Board
• Join JETIR URP
• Call For Paper
• Research Areas
• Publication Guidelines
• Sample Paper Format
• Submit Paper Online
• Processing Charges
• Hard Copy and DOI Charges
• Check Your Paper Status
• Current Issue
• Past Issues
• Special Issues
• Conference Proposal
• Recent Conference
• Published Thesis

Contact Us Click Here

Whatsapp contact click here, published in:.

Volume 7 Issue 12 December-2020 eISSN: 2349-5162

UGC and ISSN approved 7.95 impact factor UGC Approved Journal no 63975

Unique identifier.

Published Paper ID: JETIR2012317

Registration ID: 304644

Page Number

Post-publication.

• Downlaod eCertificate, Confirmation Letter
• editor board member
• JETIR front page
• Journal Back Page
• UGC Approval 14 June W.e.f of CARE List UGC Approved Journal no 63975

Share This Article

Important links:.

• Call for Paper
• Submit Manuscript online

• Shaikh Saqib
• Prof. Shaikh Mohd Ashfaque

Cite This Article

2349-5162 | Impact Factor 7.95 Calculate by Google Scholar An International Scholarly Open Access Journal, Peer-Reviewed, Refereed Journal Impact Factor 7.95 Calculate by Google Scholar and Semantic Scholar | AI-Powered Research Tool, Multidisciplinary, Monthly, Multilanguage Journal Indexing in All Major Database & Metadata, Citation Generator

Publication Details

Download paper / preview article.

Download Paper

Preview this article, download pdf, print this page.

Impact Factor:

Impact factor calculation click here current call for paper, call for paper cilck here for more info important links:.

• Follow Us on

• Developed by JETIR

Personalize Your Experience

Log in or create an account for a personalized experience based on your selected interests.

Already have an account? Log In

Free standard shipping is valid on orders of $45 or more (after promotions and discounts are applied, regular shipping rates do not qualify as part of the $45 or more) shipped to US addresses only. Not valid on previous purchases or when combined with any other promotional offers.

Register for an enhanced, personalized experience.

Receive free access to exclusive content, a personalized homepage based on your interests, and a weekly newsletter with topics of your choice.

Home / Healthy Aging / AI in healthcare: The future of patient care and health management

AI in healthcare: The future of patient care and health management

Curious about artificial intelligence? Whether you're cautious or can't wait, there is a lot to consider when AI is used in a healthcare setting.

Please login to bookmark

With the widespread media coverage in recent months, it’s likely that you’ve heard about artificial intelligence (AI) — technology that enables computers to do things that would otherwise require a human’s brain. In other words, machines can be given access to large amounts of information, and trained to solve problems, spot patterns and make recommendations. Common examples of AI in everyday life are virtual assistants like Alexa and Siri.

What you might not know is that AI has been and is being used for a variety of healthcare applications. Here’s a look at how AI can be helpful in healthcare, and what to watch for as it evolves.

What can AI technology in healthcare do for me?

A report from the National Academy of Medicine identified three potential benefits of AI in healthcare: improving outcomes for both patients and clinical teams, lowering healthcare costs, and benefitting population health.

From preventive screenings to diagnosis and treatment, AI is being used throughout the continuum of care today. Here are two examples:

Preventive care

Cancer screenings that use radiology , like a mammogram or lung cancer screening, can leverage AI to help produce results faster.

For example, in polycystic kidney disease (PKD), researchers discovered that the size of the kidneys — specifically, an attribute known as total kidney volume — correlated with how rapidly kidney function was going to decline in the future.

But assessing total kidney volume, though incredibly informative, involves analyzing dozens of kidney images, one slide after another — a laborious process that can take about 45 minutes per patient. With the innovations developed at the PKD Center at Mayo Clinic, researchers now use artificial intelligence (AI) to automate the process, generating results in a matter of seconds.

Bradley J. Erickson, M.D., Ph.D., director of Mayo Clinic’s Radiology Informatics Lab, says that AI can complete time-consuming or mundane work for radiology professionals , like tracing tumors and structures, or measuring amounts of fat and muscle. “If a computer can do that first pass, that can help us a lot,” says Dr. Erickson.

Risk assessment

In a Mayo Clinic cardiolog y study , AI successfully identified people at risk of left ventricular dysfunction, which is the medical name for a weak heart pump , even though the individuals had no noticeable symptoms. And that’s far from the only intersection of cardiology and AI.

“We have an AI model now that can incidentally say , ‘Hey, you’ve got a lot of coronary artery calcium, and you’re at high risk for a heart attack or a stroke in five or 10 years,’ ” says Bhavik Patel, M.D., M.B.A., the chief artificial intelligence officer at Mayo Clinic in Arizona.

How can AI technology advance medicine and public health?

When it comes to supporting the overall health of a population, AI can help people manage chronic illnesses themselves — think asthma, diabetes and high blood pressure — by connecting certain people with relevant screening and therapy, and reminding them to take steps in their care, such as take medication.

AI also can help promote information on disease prevention online, reaching large numbers of people quickly, and even analyze text on social media to predict outbreaks. Considering the example of a widespread public health crisis, think of how these examples might have supported people during the early stages of COVID-19. For example, a study found that internet searches for terms related to COVID-19 were correlated with actual COVID-19 cases. Here, AI could have been used to predict where an outbreak would happen, and then help officials know how to best communicate and make decisions to help stop the spread.

How can AI solutions assist in providing superior patient care?

You might think that healthcare from a computer isn’t equal to what a human can provide. That’s true in many situations, but it isn’t always the case.

Studies have shown that in some situations, AI can do a more accurate job than humans. For example, AI has done a more accurate job than current pathology methods in predicting who will survive malignant mesothelioma , which is a type of cancer that impacts the internal organs. AI is used to identify colon polyps and has been shown to improve colonoscopy accuracy and diagnose colorectal cancer as accurately as skilled endoscopists can.

In a study of a social media forum, most people asking healthcare questions preferred responses from an AI-powered chatbot over those from physicians, ranking the chatbot’s answers higher in quality and empathy. However, the researchers conducting this study emphasize that their results only suggest the value of such chatbots in answering patients’ questions, and recommend it be followed up with a more convincing study.

How can physicians use AI and machine learning in healthcare?

One of the key things that AI may be able to do to help healthcare professionals is save them time . For example:

• Keeping up with current advances. When physicians are actively participating in caring for people and other clinical duties, it can be challenging for them to keep pace with evolving technological advances that support care. AI can work with huge volumes of information — from medical journals to healthcare records — and highlight the most relevant pieces.
• Taking care of tedious work. When a healthcare professional must complete tasks like writing clinical notes or filling out forms , AI could potentially complete the task faster than traditional methods, even if revision was needed to refine the first pass AI makes.

Despite the potential for AI to save time for healthcare professionals, AI isn’t intended to replace humans . The American Medical Association commonly refers to “augmented intelligence,” which stresses the importance of AI assisting, rather than replacing, healthcare professionals. In the case of current AI applications and technology, healthcare professionals are still needed to provide:

• Clinical context for the algorithms that train AI.
• Accurate and relevant information for AI to analyze.
• Translation of AI findings to be meaningful for patients.

A helpful comparison to reiterate the collaborative nature needed between AI and humans for healthcare is that in most cases, a human pilot is still needed to fly a plane. Although technology has enabled quite a bit of automation in flying today, people are needed to make adjustments, interpret the equipment’s data, and take over in cases of emergency.

What are the drawbacks of AI in healthcare?

Despite the many exciting possibilities for AI in healthcare, there are some risks to weigh:

• If not properly trained, AI can lead to bias and discrimination. For example, if AI is trained on electronic health records, it is building only on people that can access healthcare and is perpetuating any human bias captured within the records.
• AI chatbots can generate medical advice that is misleading or false, which is why there’s a need for effectively regulating their use.

Where can AI solutions take the healthcare industry next?

As AI continues to evolve and play a more prominent role in healthcare, the need for effective regulation and use becomes more critical. That’s why Mayo Clinic is a member of Health AI Partnership, which is focused on helping healthcare organizations evaluate and implement AI effectively, equitably and safely.

In terms of the possibilities for healthcare professionals to further integrate AI, Mark D. Stegall, M.D., a transplant surgeon and researcher at Mayo Clinic in Minnesota says, “I predict AI also will become an important decision-making tool for physicians.”

Mayo Clinic hopes that AI could help create new ways to diagnose, treat, predict, prevent and cure disease. This might be achieved by:

• Selecting and matching patients with the most promising clinical trials.
• Developing and setting up remote health-monitoring devices.
• Detecting currently imperceptible conditions.
• Anticipating disease-risk years in advance.

Relevant reading

Mayo Clinic Guide to Better Vision, Third Edition

Mayo Clinic Guide to Better Vision, Sophie J. Bakri, M.D., walks readers through the diagnoses and treatment options associated with common eye issues, as well as preventive measures for protecting your eyes.

Mayo Clinic on Incontinence

This easy-to-understand book is a source of hope and assurance for all individuals who experience bladder and bowel control problems. Its message is that incontinence isn’t something people need to live with; most incontinence can be improved or cured. Mayo Clinic on Incontinence is a helpful resource to successfully manage…

Discover more Healthy Aging content from articles, podcasts, to videos.

You May Also Enjoy

Privacy Policy

We've made some updates to our Privacy Policy. Please take a moment to review.

Help | Advanced Search

Computer Science > Machine Learning

Title: mixing artificial and natural intelligence: from statistical mechanics to ai and back to turbulence.

Abstract: The paper reflects on the future role of AI in scientific research, with a special focus on turbulence studies, and examines the evolution of AI, particularly through Diffusion Models rooted in non-equilibrium statistical mechanics. It underscores the significant impact of AI on advancing reduced, Lagrangian models of turbulence through innovative use of deep neural networks. Additionally, the paper reviews various other AI applications in turbulence research and outlines potential challenges and opportunities in the concurrent advancement of AI and statistical hydrodynamics. This discussion sets the stage for a future where AI and turbulence research are intricately intertwined, leading to more profound insights and advancements in both fields.

Submission history

Access paper:.

• HTML (experimental)
• Other Formats

References & Citations

• Google Scholar
• Semantic Scholar

BibTeX formatted citation

Bibliographic and Citation Tools

Code, data and media associated with this article, recommenders and search tools.

• Institution

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .

Generative Artificial Intelligence in education: Think piece by Stefania Giannini

Artificial Intelligence tools open new horizons for education, but we urgently need to take action to ensure we integrate them into learning systems on our terms. That is the core message of UNESCO’s new paper on generative AI and the future of education . In her think piece, UNESCO Assistant Director-General for Education, Stefania Giannini expresses her concerns that the checks and balances applied to teaching materials are not being used to the implementation of generative AI. While highlighting that AI tools create new prospects for learning, she underscores that regulations can only be built once the proper research has been conducted.

Readiness of schools to regulate the use of AI tools in education

In May, a UNESCO global survey of over 450 schools and universities found that fewer than 10% have developed institutional policies and/or formal guidance concerning the use of generative AI applications. The paper observes that in most countries, the time, steps and authorizations needed to validate a new textbook far surpass those required to move generative AI utilities into schools and classrooms. Textbooks are usually evaluated for accuracy of content, age-appropriateness, relevance of teaching and accuracy of content, cultural and social suitability which encompasses checks to protect against bias, before being used in the classroom.

Education systems must set own rules

The education sector cannot rely on the corporate creators of AI to regulate its own work. To vet and validate new and complex AI applications for formal use in school, UNESCO recommends that ministries of education build their capacities in coordination with other regulatory branches of government, in particular those regulating technologies.

Potential to undermine the status of teachers and the necessity of schools

The paper underscores that education should remain a deeply human act rooted in social interaction. It recalls that during the COVID-19 pandemic, when digital technology became the primary medium for education, students suffered both academically and socially. The paper warns us that generative AI in particular has the potential to both undermine the authority and status of teachers, and to strengthen calls for further automation of education: Teacher-less schools, and school-less education. It emphasizes that well-run schools, coupled with sufficient teacher numbers, training and salaries must be prioritized.

Education spending must focus on fundamental learning objectives

The paper argues that investment in schools and teachers, is the only way to solve the problem that today, at the dawn of the AI Era, 244 million children and youth are out of school and more than 770 million people are non-literate. Evidence shows that good schools and teachers can resolve this persistent educational challenge – yet the world continues to underfund them.

UNESCO’s response to generative AI in education

UNESCO is steering the global dialogue with policy-makers, EdTech partners, academia and civil society. The first global meeting of Ministers of Education took place in May 2023 and the Organization is developing policy guidelines on the use of generative AI in education and research, as well as frameworks of AI competencies for students and teachers for school education. These will be launched during the Digital Learning Week , which will take place at UNESCO Headquarters in Paris on 4-7 September 2023. The UNESCO Global Education Monitoring Report 2023 to be published on 26 July 2023 will focus on the use of technology in education.

UNESCO’s Recommendation on the Ethics of Artificial Intelligence

UNESCO produced the first-ever global standard on AI ethics – the ‘Recommendation on the Ethics of Artificial Intelligence’ in November 2021. This framework was adopted by all 193 Member States. The Recommendation stresses that governments must ensure that AI always adheres to the principles of safety, inclusion, diversity, transparency and quality.

Related items

• Artificial intelligence
• Future of education
• Educational technology
• Digital learning week
• Educational trends
• Topics: Display
• See more add

More on this subject

Other recent idea

A comprehensive bibliometric and content analysis of artificial intelligence in language learning: tracing between the years 2017 and 2023

• Open access
• Published: 01 April 2024
• Volume 57 , article number  107 , ( 2024 )

Cite this article

You have full access to this open access article

• Abdur Rahman 1 ,
• Antony Raj 1   na1 ,
• Prajeesh Tomy 1   na1 &
• Mohamed Sahul Hameed 1   na1  

The rising pervasiveness of Artificial Intelligence (AI) has led applied linguists to combine it with language teaching and learning processes. In many cases, such implementation has significantly contributed to the field. The retrospective amount of literature dedicated on the use of AI in language learning (LL) is overwhelming. Thus, the objective of this paper is to map the existing literature on Artificial Intelligence in language learning through bibliometric and content analysis. From the Scopus database, we systematically explored, after keyword refinement, the prevailing literature of AI in LL. After excluding irrelevant articles, we conducted our study with 606 documents published between 2017 and 2023 for further investigation. This review reinforces our understanding by identifying and distilling the relationships between the content, the contributions, and the contributors. The findings of the study show a rising pattern of AI in LL. Along with the metrics of performance analysis, through VOSviewer and R studio (Biblioshiny), our findings uncovered the influential authors, institutions, countries, and the most influential documents in the field. Moreover, we identified 7 clusters and potential areas of related research through keyword analysis. In addition to the bibliographic details, this review aims to elucidate the content of the field. NVivo 14 and Atlas AI were used to perform content analysis to categorize and present the type of AI used in language learning, Language learning factors, and its participants.

Avoid common mistakes on your manuscript.

1 Introduction

Artificial Intelligence (AI) holds a pivotal role in rebuilding our society and ‘promising a new era’ in prospective times for its capabilities to act as intelligent beings in various domains including education (Farrokhnia et al. 2023 ; Górriz et al. 2020 ). An upsurge, in recent times, on the application of AI in educational sectors, has exhibited notable development, and there has been an equivalent explosion in the number of new AI tools accessible (Chu et al. 2022 ; Popenici and Kerr 2017 ). In the field of education, researchers report on the opportunities that AI presents for instructors and learners (Chen et al. 2020 ). This evolutionary trajectory of AI in education is increasing, showcasing an exponential growth of its explorations across disciplines including language education (Jeon et al. 2023 ).

The use of AI, in particular with language learning, is appreciated for providing the students with individualized attention, “personalized, interactive, and authentic language learning contexts” in the form of intelligent tools such as Interactive Personal Assistants, web-based systems, virtual reality systems, or chatbots (Lin and Chang 2020 ; Liang et al. 2021 ; Wijekumar et al. 2013 ; Rahman and Tomy 2023 ; Zhang et al. 2023 ). Moreover, it allows the teachers/instructors to monitor their students/learners, which reduces their workload and frees the teachers thus allowing them to prioritize the curriculum over repetitive tasks (Pokrivčáková 2019 ). Integration of AI techniques such as natural language processing (NLP), natural language understanding (NLU), and automatic speech recognition (ASR) allows the use of tools developed through them to be more appropriate in language learning platforms as they comprehend and process human-computer interaction (Lee and Jeon 2022 ; Shadiev and Liu 2022 ).

While multiple studies address the research gaps in using AI in language learning, the advancement of AI is at a much faster phase extending the research need perpetually. Gruetzemacher ( 2022 ) states that “In the past two years, Language-based AI has advanced by leaps and bounds, changing common notions of what this technology can do”. The technological impediments of the past decades are no longer the present-day concerns. Therefore, the research trends and findings of previous studies slither on every massive leap of AI techniques. For example, in 2017, advancements in word embedding (Peters et al. 2017 ), and the introduction of Transformer model architecture in the field of NLP outperformed prior recurrent neural network models and offered a novel technique to carry out sequence transduction tasks (Vaswani et al. 2017 ). Later, the launch of ChatGPT and GPT-4 in 2022 and 2023 offered remarkable conversational capabilities and longer text processing demonstrating keen anticipation for NLPs and the next phase of human-computer interaction (Weitzman 2023 ). Such advancements of NLP greatly help intelligent systems uncover the unstructured data produced by humans.

The production and reception of natural language consequently expand the use of AI for language learning. Past studies have examined these phenomena of AI in language learning and education. For instance, Liang et al. ( 2021 ) performed a bibliographic analysis and systematic review on 71 articles on AILEd (AI in language education) on December 31, 2020. Even though research on language teaching and learning in connection with AI is active, given the dynamic progressive nature of AI techniques, reviewing their implications and applications on language learning, within the designated timeframe, could considerably contribute to the field. Hence, it is essential to state the art of AI in language learning at timely intervals. To review the state-of-art of AI in language learning, a bibliometric analysis was performed. This analysis could segment a “large volume of scientific publications” with the “advancement, availability, and accessibility of bibliometric software and scientific databases”, (Donthu et al. 2021 ). It could objectively point out the performance and emerging trends in the given field including topics and authors (Verma and Gustafsson 2020 ). With the study’s primary objective being to review the developments in academic research on AI and language learning between the years 2017 and 2023, the study focuses (1) to analyze the publication trends and growth patterns of AI in language learning (2) to identify the key contributors, collaboration patterns, and influential works in the field (3) to explore the dominant research themes and emerging trends. To operationalize the objectives of the research, they were converted into the following research questions to identify the mentioned publication metrics.

1.1 Research questions

RQ1: What are the publication trends and metrics of performance analysis such as Publication, Citation, and both Citation and publication-related metrics?

RQ2: Who or Which are the most influential and productive authors, institutions, journals, and countries?

RQ3: What are the key research themes, frequent and prominent keywords obtained from title, abstract, and keywords through keywords analysis?

RQ4: What are the documents and clusters that are connected to a common document’s reference through bibliographic coupling?

RQ5: what are the inferences obtained by analyzing the content of all the documents in the study through content analysis? With the study’s primary objective being to review the developments in academic research on AI and language learning between the years 2017 and 2023, the study focuses (1) to analyze the publication trends and growth patterns of AI in language learning (2) to identify the key contributors, bibliographic clusters, and influential works in the field (3) to explore the dominant research themes and emerging trends. To operationalize the objectives of the research, they were converted into research questions to identify the mentioned publication metrics. (RQ1) provides valuable insights into the growth and impact of this interdisciplinary domain. This information can guide researchers, policymakers, and educators in identifying areas of prominence and potential gaps in the literature. (RQ2) allows us to recognize key contributors to the field and potentially foster collaborations. Additionally, (RQ3) helps in analyzing key research themes and prominent keywords to comprehend the evolving discourse and focus of research in this area. Moreover, exploring document connections through bibliographic coupling (RQ4) aids in mapping the intellectual structure of the field. Lastly, extracting inferences through content analysis (RQ5) offers insights into the practical implications of the existing research, potentially informing pedagogical tools used, their frequency of usage, the target learners and also the language learning factors. In sum, addressing these research questions not only contributes to the scholarly understanding of this domain but also has practical implications for educators, researchers, and stakeholders invested in the intersection of AI and language learning.

2 Background of the study

2.1 artificial intelligence.

While multiple researchers have laid out technical definitions to bind AI within a school of thought, Russell and Norvig ( 2010 ) categorizes these definitions under two dimensions. First, it has the ability to imitate, think and act humanly and rationally. Second, its connection with the thought process and reasoning along with its behavior. In a broader context, AI involves computing technology that allows machines to mimic human intelligence “in analysis, reasoning, decision making, and self-correction” (Liang et al. 2021 ; Pokrivčáková 2019 ). To perform the above-mentioned operations, a wide variety of techniques and methods are used such as “machine learning, adaptive learning, natural language processing, data mining, crowdsourcing, neural networks or an algorithm” (Pokrivčáková 2019 ).

Despite its complicated mechanisms and progressive developments, IBM (n.d) states that “there is no practical examples of strong AI in use today”. However, as given in Table  1 , AI has had many technological breakthroughs from computer vision to advanced natural language processing techniques over a brief period and it is, in many cases, being considered a substitute for human intelligence. At its present rate of growth, AI is presumed to surpass human intelligence. Thus, a review of its applications in various domains is a pressing priority.

2.2 Integration of artificial intelligence in language learning

The incorporation of AI in any field can be in multiple technological forms, tools, or software (Thayyib et al. 2023 ). Similarly, in language learning and acquisition, a diverse assortment of tools is being integrated with artificial intelligence as it offers language teachers and learners “personalized, interactive and adaptive learning experiences that cater to individual’ needs and preferences” (Rusmiyanto et al. 2023 ; Pokrivčáková 2019 ). Research is being carried out to identify appropriate AI-assisted tools to improve each language skill (Rahman et al. 2022 ) and on the integration of each tool to develop specific areas of language learning and acquisition.

The possibilities of advanced technological input through AI have made Computer Assisted Language Learning (CALL) conventional. Researchers argued that CALL had been limited when proposed with activities directed towards communication and interaction between students such as role plays, discussions, and sharing opinions (Amaral and Meurers 2011 ). The notion of computers replacing humans in CALL was skeptical as language learning reinforces negotiating meaning and having real-time conversations. Later, the increased possibility of human-like interaction with computers through AI techniques, made Intelligent Computer Assisted Language Learning (ICALL) gain its state as a potential computing technology that reformed CALL through its dynamics in multiple aspects (Segler et al. 2002 ; Esit 2011 ; Amaral and Meurers 2011 ). However, not all AI-based tools are directly associated with ICALL, they are diversified into multiple forms and tools as follows; Adaptive Educational System (Triantafillou et al. 2003 ), Intelligent Educational System (Cumming et al. 1993 ) Intelligent Personal Assistant (Rahman and Tomy 2023 ; Yang et al. 2022 ) Intelligent Tutoring System (Slavuj et al. 2015 ), Natural language processing (Nagata 2013 ), Machine translation tools (Briggs 2018 ), Chatbots (Jeon 2021 ; Dokukina and Gumanova 2020 ), AI writing assistants (Gayed et al. 2022 ; Godwin-Jones 2022 ), AI-powered language learning software (Pokrivčáková 2019 ) and Intelligent Virtual Reality (Ma 2021 ). These AI-powered language learning tools are used to foster students’ language skills and sub-skills, encourage students’ interaction, reduce the affective factors of language learning and acquisition, push their willingness to communicate, and more (Tai and Chen 2020 ; Shazly 2021 ; Liang et al. 2021 ). Acknowledging its efficient performance in multiple studies, researchers and applied linguists are paying attention to AI-empowered tools and their applications.

2.3 Previous reviews on AI in language learning and education

Like given in Table  2 , previous review papers have analyzed the trends and research foci of artificial intelligence and AI-powered tools in language learning and education. Review papers have focused on specified AI-based tools like Intelligent tutoring systems, Voice based virtual agents and speech recognition chatbots (Xu et al. 2019 ; Katsarou et al. 2023 ; Jeon et al. 2023 ) and have worked on reviewing the integration of intelligent tools such as ChatGPT and conversational AI, and approaches like ICALL in language learning (Kohnke et al. 2023 ; Weng and Chiu 2023 ; Ji et al. 2022 ). Likewise, researchers have reviewed the role of AI in language learning and language education(Huang et al. 2021 ; Liang et al. 2021 ; Fang et al. 2023 ; Ali 2020 ; Sharadgah and Sa’di 2022 ; Yang et al. 2022 ). Although studies have focused on AI and its applications, reviews on AI, in many cases, have focused either on “Artificial Intelligence” or on other applications such as “Intelligent Tutoring Systems” but not on both. Studies analyzing both aspects are notably limited. Thus, this paper aims to perform an analysis of both aspects which includes AI and AI-powered tools and approaches. Additionally, following the introduction of advanced NLP models in 2022, there has been a notable absence of dedicated reviews concerning the role of AI in language learning. Thus, this study aims to analyze the trends and impact of AI in language learning up to the year 2023. This includes the most advanced ChatGPT and other NLP-powered intelligent agents. Along with bibliometric analysis, through content analysis we also investigated its connectives with language learning factors and the target learners through the mode of instruction.

3 Methodology

3.1 defining aims and scope.

It is essential to set clear objectives and parameters before moving on to the process of bibliometric analysis Belmonte et al. ( 2020 ); Donthu et al. ( 2021 ). The authors sought to analyze the conceptual framework and reflect on influential research contributors and their collaboration in the research area. Any area of research with more than 500 papers “deserves a bibliometric analysis” (Hou and Yu 2023 ; Donthu et al. 2021 ). The scope of the study, in accordance with the regular standards, will analyze more than 500 papers. Along with bibliometric analysis, a content analysis will be performed to identify the key applications and their participants in the research area.

PRISMA method procedure for screening and selecting the documents

3.2 Data source

The authors opted for the Scopus database to carry out the bibliometric and content analysis, in line with other prior bibliometric research (Thayyib et al. 2023 ; Ahmed et al. 2022 ; Goodell et al. 2021 ). Scopus is one of the largest databases of scholarly works with more than 84 million records and 1.8 billion cited references (Home https://hai.stanford.edu/ ). With its advanced search capabilities and wide coverage, it holds the records of research works being published even in developing countries. It offers adequate bibliometric details such as citation information, bibliographic information, abstracts, keywords, funding details, and other information including references. In addition, Scopus provides these data in multiple formats to feed into software that is used to systematically analyze documents.

3.3 Data collection and refinement

The authors retrieved the data used for analysis from the Scopus database on June 22, 2023. The period range was limited between the years of 2017 and 2023 to obtain scholarly works that mostly reflect advanced AI techniques in language learning. With reference to previous relevant bibliometric and systematic reviews, the search keywords were finalized (Hou and Yu 2023 ; Liang et al. 2021 ; Popenici and Kerr 2017 ; Chu et al. 2022 ; Jeon et al. 2023 ; Tan et al. 2022 ). To further extend the scope of the study, ChatGPT was also included. However, we excluded “machine learning”, “deep learning” and “deep neural networks”. While undoubtedly, these components are important in the broader field of AI, these terms tend to yield a substantial number of papers related to computer language learning and programming languages, which are distinct from our primary focus on second language learning.

The chosen keywords were influenced by the prominent pedagogical viewpoint within second language learning. NLP techniques, conversational systems, and interactive chatbots are frequently employed in this context to facilitate meaningful interactions between the AI and the learner. Considering these factors, the following keywords were used to search relevant articles in the database ( TITLE-ABS-KEY ( "Chatbot*" OR "conversational agent" OR "pedagogical agent" OR "conversational system" OR "dialogue system" OR "spoken dialogue system" OR "intelligent personal assistant" OR "ICALL" OR "intelligent computer assisted language learning" OR "artificial intelligence" OR "intelligent tutoring system" OR "ChatGPT" OR "ChatGPT-4" OR "natural language processing" OR "NLP" ) AND TITLE-ABS-KEY ( "Language learning" OR "language teaching" OR "language acquisition" OR "second language learning" OR "foreign language learning" ) ). A total of 1870 results were obtained from the keyword search. The publication selection procedure is given in Fig.  1 Page et al. ( 2021 ), and the inclusion and exclusion criteria are in Table  3 . On the exclusion of articles based on the criteria, only 606 documents were processed for bibliometric and content analysis.

3.4 Technical tools and procedure of data analysis

Bibliometric analysis and content analysis were performed to gain an overall understanding of the research on AI in language learning. The study employed bibliometric analysis to identify the publication trends, leading authors, institutions, prominent journals, collaboration patterns, citation analysis, geographic distribution, keywords co-occurrence analysis, co-authorship analysis, and co-citation analysis. The authors used VOSviewer, Publish or Perish software, and Scopus to visualize and extract results from the retrieved data with the objective to carry out bibliometric analysis.

The authors, through content analysis, opted to identify the overview of the types of AI tools used in language learning, the language skills it is being tested against, and the educational levels of the participants of the study. After identifying the keywords through text-based content analysis on VOSviewer, a conceptual content analysis was performed, adhering to the deductive coding approach following the structural coding method through code categorization (Krippendorff 2018 ). The coding scheme for content analysis was done with NVivo, which assists in “classifying, sorting and modelling qualitative data” (Bazeley 2019 ). The schemes of the documents were initially classified through auto-coding in NVivo. In addition, manual classification was done independently by two research scholars to examine the auto-coded results. The auto-coding was fed into Atlas AI to identify the connections between the variables of the study.

4.1 Publication trends and performance analysis of AI in language learning

The screened data had 606 documents published between the years 2017 and 2023. It included 230 research articles, 29 book chapters, 330 conference papers, and 17 reviews. Among these documents, 39 were articles-in-press. Most of these documents were closed access, only 185 articles were open access among which 117 were research articles, 61 were conference papers, and 7 were reviews. However, along with the increase in the total number of documents published over time, open-access documents got doubled between 2017 and 2023. As shown in Fig.  2 , there is a noticeable growth in the total number of documents published from 2017 to 2023 in the subject area. From 2017 to 2022, there has been a gradual increase in the number of documents that were published, indicating a growth of 189.8 \(\%\) . The data presented for 2023 is not complete as it was collected during the middle of the year (June 22, 2023). However, the reported number of 72 documents signifies promising growth. While the number of research publications has increased, the number of citations has decreased over time as presented in Fig.  2 .

Publications and citations between 2017 and 2023

“Performance analysis examines the contribution of research constituents to a given field” (Donthu et al. 2021 ). With total publication and total citation details, the analysis has other metrics to be evaluated including ‘scientific actors’ like h-index and i-index (Cobo et al. 2011 ). Table  4 indicates the overall performance of AI in language learning through selected metrics from Donthu et al. ( 2021 ).

Donthu et al. ( 2021 ) states that it is a “standard practice” to present the background or profile of the retrieved documents. Thus, the study further elaborates on the contributions of the (1) authors, (2) institutions, (3) sources, and (4) countries that are highly influential in the field of AI in language learning. The following formulas were used to calculate the metrics identified in Table  4 : PAY = (TP \(\div\) NAY), ACP = TC \(\div\) TP, PCP = (NCP / TP) * 100, and CCP = (TC \(\div\) NCP). NCA was calculated by identifying (The total number of authors - Duplicates) in Microsoft Excel. NAY is the total number of years that the research constituent records the publications and NCP was identified by filtering the publications with citation in Microsoft Excel. The h-index and I- Index were calculated through Publish or Perish software by the RIS format derived from Scopus (Table 4 ).

4.2 Top authors, sources, countries, and institutions

We used VOSviewer and R studio (Biblioshiny) to identify and cross-validate individual influential authors through co-citation analysis. As given in Table  5 , the authors identified are Meurers d.; Fryer I.K.; Hwang g.j.; Dizon. G.; Chen x.; Zou d.; Strik. H; Cucchiarini c.; Thompson A.; and Xie h. In Table  6 , the top ten cited sources with a minimum of 5 documents in the field are listed. The sources are Computer Assisted Language Learning, Interactive Learning Environments, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Educational Technology and Society, Procedia Computer Science, ACM International Conference Proceeding Series, Journal of Physics: Conference Series, Proceedings of the 13th Workshop on Innovative Use of NLP for Building Educational Applications, BEA 2018 at the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL-HTL 2018, International Journal of Emerging Technologies in Learning, Applied Sciences (Switzerland). The document type of sources are Journals, Book series, and Conference proceedings. Table  7 includes documents and citations of the top ten organizations with high impact in the field of AI in language learning between the years 2017 and 2023. The University of Sydney, Himeji Dokkyō University, National Taiwan Normal University, University of Piraeus, Georgia State University, National Taiwan University of Science and Technology, University of Tübingen, The Education University of Hong Kong, Lingnan University and the University of Cambridge are the most influential organisations.

Influential countries

Figure  3 presents the top ten influential countries in the field along with their total number of documents produced and citations achieved. We found out that the United States, Japan, China, Hong Kong, Taiwan, Canada, Germany, Australia, the United Kingdom, and Indonesia are the most productive and influential countries. VOSviewer was used to identify these countries by evaluating the citations against countries. The minimum number of documents per country was set to 10 to identify countries that are both productive and influential.

4.3 Keywords analysis

Co-occurrence of keywords visualization

The co-occurrences of keywords analysis was made through VOSviewer to identify the keywords that authors use frequently. With full counting, the co-occurrence of the author keywords has opted with keyword occurrences of \(>5\) . Of the 3252 keywords, only 184 (5.66 \(\%\) ) met the threshold with 4069 links and a total link strength of 10134. The overall strength of the co-occurrence connections between each of the 184 keywords was calculated. By doing so, Fig.  4 was generated.

The 184 keywords were classified into seven clusters. Table 8 details the seven clusters. Cluster 1 consists of 49 items, like language learning (f = 193, Total link strength = 1026), Natural language processing systems (f = 132, Total link strength = 817), Computational linguistics (f = 46, Total link strength = 262), Second language acquisition (f = 44, Total link strength = 207), and Deep Learning (f = 32, Total link strength = 231). Cluster 2 comprised 43 items, including Artificial Intelligence (f = 211, Total link strength = 1051), Students (f = 112, Total link strength = 784), Teaching (f = 94, Total link strength = 664), Engineering Education (f = 49, Total link strength = 397), and Education computing (f = 39, Total link strength = 317). Cluster 3 has 43 items and the most occurred keywords in this cluster are E-learning (f = 78, Total link strength = 556), linguistics (f = 39, Total link strength = 283), Education (f = 27, Total link strength = 175), Chatbot (f = 26, Total link strength = 92), and English languages (f = 24, Total link strength = 183). Cluster 4 consists of 16 items, including Learning systems (f = 125, Total link strength = 896), Foreign language (f = 36, Total link strength = 238), Computer Assisted Language Learning (f = 28, Total link strength = 195), Intelligent computer-assisted language learning (f = 16, Total link strength = 61), and Error correction (f = 10, Total link strength = 62). 14 items were found in cluster 5, like Computer aided instruction (f = 109, Total link strength = 824), Intelligent tutoring systems (f = 25, Total link strength = 178), Foreign language learning (f = 24, Total link strength = 142), Educational technology (f = 13, Total link strength = 79), and Tutoring system (f = 10, Total link strength = 85). Cluster 6 comprises 11 items including Speech recognition (f = 24, Total link strength = 181), Automatic speech recognition (f = 8, Total link strength = 70), Machine translations (f = 9, Total link strength = 62), Computer-aided language translation (f = 7, Total link strength = 57), and Deep neural networks (f = 7, Total link strength = 48). In Cluster 7, 8 items were found including Teacher (f = 27, Total link strength = 212), Learning platform (f = 8, Total link strength = 48), Decision making (f = 6, Total link strength = 44), Statistical tests (f = 5, Total link strength = 35), and online learning (f = 5, Total link strength = 32).

4.4 Bibliographic coupling

Bibliographic coupling based on Documents

Bibliographic coupling was done to identify literature that is connected through a common document’s reference. The network visualization of the documents as illustrated in Fig.  5 details the interconnections between the documents through 5 clusters. To perform this visualisation, full counting was opted with the unit of analysis as Bibliometric coupling with documents. In order to narrow down the influential works, the minimum number of citations per document was set to 20. Of the 606 documents, only 40 met the threshold. Among the filtered 40 documents, only 21 items in the network had the largest set of connected items.The most influential authors of AI in language learning are listed in Table  9 .

The most influential documents identified through Bibliometric coupling are “Stimulating and sustaining interest in a Language Course: An experimental comparison of Chatbot and Human task partners” (Fryer et al. 2017 ), “Technology and the Future of language teaching” (Kessler 2018 ), “Chatbot learning partners: Connecting learning experiences, interest, and competence” (Fryer et al. 2019 ), and “Using Intelligent Personal Assistants for Second Language Learning: A Case Study of Alexa” Dizon ( 2017 ), and “Chatbots for language learning-Are they really useful? A systematic review of chatbot-supported language learning” (Huang et al. 2021 ).

4.5 Content analysis

AI tools and techniques

The title and the abstract fields of all the documents were fed into VOSviewer to trace out the most occurred words. With a minimum number of occurrences per term set to 10, we identified 382 terms. For each of the 382 items, a relevance score was calculated by default in VOSviewer, and only 60 \(\%\) of the most relevant terms were opted for further analysis. Upon filtration, we created a map based on textual data with 299 items under 6 clusters. Among these 299 items from VOSviewer, we excluded terms that do not add up to any contextual meaning such as research gap, participants, perception, English language teaching, methodology, experiment results, observation, sample, control group, experimental group, the current study, post-test, and questionnaire. A total number of 254 were excluded on such pretest. A total number of 45 terms were manually coded through NVivo 14 into three different schemes or parent codes namely (1) Artificial Intelligence Tools and Techniques (2) Participants (3) Language Learning Factors.

AI tools and techniques gave us an overview on the type of AI-based technology that is implemented in the studies, the participants’ parent code had the different age groups of learners upon whom the experiments had been conducted, and the Language Learning Factors parents code had the factors with which the AI techniques were tested against. Figure  6 illustrates the 45 items under three different clusters. The child codes are organized under parent codes. The terms used most frequently under AI tools and techniques are as follows with their occurrence in references: AI Chatbots (f = 521), ChatGPT (f = 22), Conversational Agent (f = 217), Automatic Speech Recognition (f = 216), Intelligent Personal Assistants (f = 216), Google Assistant (f = 46), Amazon Alexa (f = 24), Agent (f = 209), Virtual Reality (f = 171), Natural Language Processing (f = 170), CALL (f = 152), ICALL (f = 79), Machine Translation (f = 134), Web (f = 108), Application (f = 94), Intelligent Tutoring System (f = 60), Robot (f = 45), Error Correction (f = 43), Gengobot (f = 40), MALL (f = 39), Mobile Devices (f = 15), Mobile Learning (f = 12), Mobile Applications (f = 8), ICT (f = 29), and Gamification (f = 28). The terms most frequently used under the parent code Language Learning Factors are Writing (f = 184), Speaking (f = 153), Vocabulary (f = 148), Listening (f = 106), Grammar (f = 92), Proficiency (f = 81), Accuracy (f = 79), Pronunciation (f = 71), Fluency (f = 66), Motivation (f = 64), Sentence pattern (f = 61), Comprehension (f = 57), Anxiety (f = 32), and Formulaic sequence (f = 22). Third, the terms most frequently used under the parent code of participants are University students (f = 116), Children (f = 100), College students (f = 85), Language teachers (f = 65), Higher education (f = 63), and School students (f = 18).

Two field plots of the relationships

Then, a code-occurrence analysis was conducted to present a two-field plot to depict the relationship between the participants and the AI tools and techniques used and between language learning factors and AI tools and techniques. In order to do that, we fed the parent and child codes into ATLAS AI to generate the visualizations in Fig.  7 and to identify the link strength between the parent and its sub-codes. The plot showcases the co-occurrence patterns between the two variables. The dense clusters illustrated in Fig.  7 b indicate strong link strength such as the link between Teachers and chatbots and between chatbots and grammar in Fig.  7 a. Though it depicts the association between the tested variables, not all the variables that were fed were displayed due to their weak association such as ChatGPT in Fig.  7 a and formulaic sequences in Fig.  7 b.

5 Discussion

RQ1: What are the publication trends and metrics of performance analysis such as Publication, Citation, and both Citation and Publication-related metrics?

RQ1 is devoted to identifying the research trend of AI in language learning. The annual total publication and citation records could provide an overview of the future of the research area and its potential. The findings of publication trends given in Fig.  2 depict a gradual growth in terms of production (publication) till 2022. As the data was collected on June 22, 2023, the production rate is still incomplete. However, a promising amount of literature has been produced within the first half of the year. In contrast to the rising publication rates, a decrease in citation records could be observed. The decrease in citations may have been attributed to the research focus shift. Over time, researchers have been exploring new AI-based technologies that have not gained much attention. Researchers have shifted focus from generic terms such as ”Artificial Intelligence” to specific tools such as IPA and ChatGPT. In both cases, the need for researchers to cite other specific applications and tools is low. The plausible reason for the reduction in citations could be the saturation of the field. The growth may have reached a point where new papers are not cited as frequently as old papers which are considered to be foundational works. We further analyzed various other metrics of publication and citation to gain insights into AI in LL. The findings shed light on productivity, its impact, and the rate of collaboration in the field. With the total number of included publications, sole-authored publications, and co-authored publications, we evaluated the level of collaboration among the researchers. The results revealed that the ( CI = 0.76) indicated a strong culture of collaboration among the researchers. The productivity of AI in language learning (TP = 606; PAY = 87) is on par with other renowned bibliometric or systematic reviews on pedagogic techniques in language learning like Virtual tools with (TP = 104), Mobile assisted vocabulary learning with (TP = 687), synchronous computer-mediated communication with (TP = 1292), and Augmented Reality with (TP = 1275) despite excluding documents published before 2017 (Botero-Gomez et al. 2023 ; Daǧdeler 2023 ; Hou and Yu 2023 ; Min and Yu 2023 ).

The research output of AI in LL has received a total of 3194 citations, with an average of 5.27 citations per document and an ACY of 456.28, indicating that the academic community has a positive reception of the produced documents. Other citation metrics such as CCP and PCP evaluate the amount and the impact of the influential works of the field. Notably, more than 60 \(\%\) of the documents had citation records with an average CCP of 8.72, suggesting a high percentage of influential papers in the field. We further examined the research impact indices to reflect the overall impact of authors in the field. The (h-index = 27, G = index = 41, and I10 index = 85) for our dataset suggests that the scholars in our field have had a significant impact.

The second research question aims at identifying the top authors, institutions, countries, and journals. The results of the analysis provided valuable insights into the contributors to the field. Our study revealed prolific authors who have made a major contribution to the field of AI in LL. The identified authors have contributed 30 documents altogether with a strong collaboration pattern between each other and other authors of the field. Only 10 \(\%\) of the 30 documents were sole-authored publications, and 90 \(\%\) being collaborative contributions. Moreover, it was found that 30 \(\%\) of the documents produced by these authors demonstrated collaboration among themselves.

The authors focused mostly on discussing the general trends and problems surrounding AI (Huang et al. 2023 ; Chen et al. 2021 ; Liang et al. 2021 ) and its tools such as Chatbots (Zhang et al. 2023 ; Huang et al. 2021 ; Fryer et al. 2019 ), ChatGPT (Kohnke et al. 2023 ), IPAs like Alexa and Google Assistant (Dizon et al. 2022 ; Dizon 2021 ), ICALL (Chen et al. 2022 ; Ruiz et al. 2019 ), Grammarly (Dizon 2021 ), Natural Language Processing (Ziegler et al. 2017 ), Speech (Litman et al. 2018 ) and digital technologies (Liu et al. 2023 ; Kienberger et al. 2022 ). While influential authors play an indelible mark in any field, equally remarkable is the role of academic sources who support the research fields. In the field of AI in LL, the most influential sources are journals (5) followed by conference proceedings (4) and book series (1). The journals have produced ( \({\bar{x}}\) = 8.6, SD = 3.36, Min = 5, Max = 13) documents with the impact measured through citation of ( \({\bar{x}}\) = 93.2, SD = 50.74, Min = 30, Max = 166) between 2017 and 2023. The renowned journals of the field publish articles about AI in LL under the categories of language and linguistics, computer science applications, and education which indicates that the research field is multidisciplinary and not bound to any particular school of thought.

In accordance with the knowledge gained from influential journals, organizations that publish influential works include not only the Department of English Language Education but also interdisciplinary departments such as Institute of Technology, Institutes of Digital Learning and Education, Department of Mathematics and Information Technology, Department of informatics, Computer Science Department, Institutes of Automated Language Teaching and Assessment, and Institute and Department of Computer Science and Technology. Thus, maintaining a highly multidisciplinary approach in the field. On the other hand, when we looked at the most productive and impactful countries in the field, we found that the majority of 156 documents on AI came from China, and the United States has got the highest number of citations of 778 with 78 documents. The results of our study are consistent with other bibliometric studies that link AI with other sectors, such as Big Data Analytics (Thayyib et al. 2023 ), Food Safety (Liu et al. 2023 ), and Smart Buildings (Luo 2022 ), although a similar study on the role of AI in language education claimed that Taiwan, the US, and the United Kingdom had secluded the top most spots. The prior analysis by Liang et al. ( 2021 ) examined documents between 1889 and 2020, whereas our study looked at documents published between 2017 and 2023, which may have led to a difference in our results. Thus, the overall analysis of RQ2 provides insights into the most influential authors, institutions, sources, and countries which can guide researchers to understand the factors that contribute to their success.

RQ3: What are the key research themes, frequent and prominent keywords obtained from title, abstract, and keywords through keyword analysis?

RQ3 illustrated extensively used keywords of AI in LL. The authors merely listed the keywords that were automatically retrieved and clustered by VOSviewer. The list contains highly occurred keywords with strong TLS and occurrences. Results reveal that “Natural language processing systems” is the most occurred technical keyword apart from “Artificial Intelligence”. However, contextual meaning or research inferences could not be obtained through the use of NLP in language learning as most AI-based systems used in language learning and acquisition platforms uses tools that are incorporated with NLP (Meurers 2012 ; Zilio et al. 2017 ). But hints for future studies could be obtained through keywords that have lower TLS and occurrences. The identified keywords with weaker connections could be focused by the researchers if found to be potential areas of research.

In RQ4, documents that were often cited by other authors of the same field were identified through bibliographic coupling. The top documents identified through bibliometric coupling were published in the year 2017 followed by 2018 and 2020. According to Dogan et al. ( 2023 ), a significant amount of literature was produced on AI in education in 2018. Our study, which aligns closely with Dogan’s findings, also observed a similar pattern, with a high number of influential works published in 2017, followed by another peak in 2018. The use of AI, Chatbots, and Alexa are discussed in most documents (Fryer et al. 2017 ; Dizon 2017 ; Huang et al. 2021 ). These documents are seen, in most cases, to be foundational works, which could be the cause of the declining citation patterns as discussed in RQ1. The inferences obtained through bibliometric coupling identify the key papers and shed light on the research landscape.

Hierarchical chart of the content analysis through schematic coding

RQ5: what are the inferences obtained by analyzing the content of all the documents in the study through content analysis?

RQ5 aimed at quantitatively contextualizing the content by coding the documents into clusters and therefore deducing inferences. Figure  8 illustrates the types of AI used in language learning, the language learning factors, and its participants based on hierarchy compared by a number of coding references. According to the model given, a large number of studies have been conducted with AI-embedded Chatbots. In line with the aforementioned statement, Jeon et al. ( 2023 ) conducted a systematic review of chatbots acknowledging their widespread application. Like Chatbots, other AI tools like CALL, Conversational agents, Virtual Reality, NLP tools, and IPAs are prevalent in the field. On the other hand, writing is the language learning factor that is mostly preferred with AI applications in language studies followed by speaking, vocabulary, proficiency, accuracy, pronunciation, listening, and fluency.

The participants that are most sought after for implying AI are university students followed by children, college students, language teachers, and students of higher education. In addition to figuring out the dominant components within the variables, we examined the interconnections among them to establish previously established research areas and research gaps. Chatbot was experimented extensively with teachers, children, and university students. Figure  7 b depicts the relationship between AI tools and the participants. Even though a high amount of interconnections could be observed with different language learning factors and AI tools, the interconnections between AI tools and the levels of participants are weak. Future studies could work on experimenting with AI tools with different participant levels. Despite the fact that there is a lot of literature on writing skills, many AI tools have been tested with speaking skills. Researchers could contribute to the field by working on weaker connections. For instance, students in colleges, universities, and schools might be exposed to different AI tools. The same could be done for fluency and anxiety, which are core areas of research with weaker connectives.

6 Conclusion

This study used bibliometric and content analysis to analyze the research trends, patterns, key contributors and content in the field of AI in LL. It summarises the bibliometric information of the field along with prominent authors, institutions, sources, and countries. A rise in publication trends has been identified. Researchers who integrate AI into language learning use a variety of tools, leading to the formation of new fields within the field and new branches within AI-based language learning. This, in turn, is speculated to be a major reason for the decline in citation trends. However, the constructive viewpoint regarding this aspect is that the researchers, utilizing diverse AI-based tools, are expected to contribute significantly to the field. Affirming this, documents that were published during 2017 and 2018 are identified, through bibliographic coupling, to be ‘often cited’ papers indicating their mark as foundational works. On analysing the bibliographic and textual data on multiple aspects, we yielded the following results:

Between 2017 and 2022, there is a considerable increase in the number of publications on AI in language learning of 189.8 \(\%\) demonstrating a promising growth in the field with 60.3 \(\%\) of the documents with citations of ( \(\ge 1\) ).

The field exhibits a significant number of co-authored publications, totalling to 466, in contrast to the relatively lower count of sole-authored publications, which stands at only 140.

As the field is emerging, a lot of new tools and technologies are being incorporated into the field. Resulting in a high number of citations for the works published in 2017 and 2018. The articles published during this period are often cited and considered as foundational works.

Our findings have identified the United States, China, and Japan, sequentially, as the most influential countries in publishing research related to AI in language learning.

On analysing the author’s keywords, we identified that there is an upsurge in the following in areas of study in connection with AI in language learning: natural language processing, computational linguistics, deep learning, speech recognition, machine translation, and deep neural networks. Among these keywords, “Natural language processing” is the most used keyword indicating the presence of Large Language Models of AI being frequently opted in language studies.

Fryer et al. ( 2017 ) and Kessler ( 2018 ), which examine the usage of chatbots in language learning and the impact of technology including AI in language learning, consequently, were shown to be the most influential texts. The first document talks about the usage of Chatbots in language learning setup, and the second document discusses the extensive use of technology in language learning. This conclusion, through bibliographic coupling, is consistent with the outcomes of the content analysis.

Through content analysis, we identified the most occurring textual terms used in the retrieved data from Scopus. We identified that the most occurring AI tool was Chatbots followed by Chatgpt, Conversational Agents, Automatic Speech Recognition and Intelligent Personal Assistants. We also identified the most researched language learning factor with AI which is Writing followed by Speaking, Vocabulary, Listening and Grammar. The most targeted participants are University Students followed by children, college students and language teachers.

6.1 Implications and contributions

In light of the rapid pace of technological advancements, several reviews are limited to incorporating the latest NLP tools, such as ChatGPT and Intelligent Personal Assistants, into their analyses. While prior bibliometric analyses have provided us with a comprehensive understanding of the bibliometric landscape concerning AI in language learning, the dynamic nature of AI necessitates an investigation inclusive of the recently launched tools, particularly in light of the technological developments emerging post-2019, with ChatGPT serving as a prime illustration thereof. Surprisingly, no prior bibliometric analysis has embraced these cutting-edge NLP tools and techniques. Furthermore, there exists a conspicuous absence of content analysis within the domain of AI in language learning. Hence, our study aims to bridge these critical gaps by providing a thorough examination of the wide range of tools and techniques utilized in AI for language learning, their respective frequencies, and the target participant groups they have been applied to. The outcomes of our research will enable future researchers to identify research gaps through content analysis by providing them with a comprehensive understanding of bibliometric information. The frequency of research among the three factors of content analysis will also serve as a vital resource for pinpointing areas that needs research attention.

6.2 Limitation of the study and recommendation for future studies

Through addressing the limitation of the study, we would want to suggest areas for further research. First, the study is limited to the Scopus database and between the years 2017 and 2023. Despite Scopus being an academically promising database for language studies, documents published on the Web of Science, ERIC, ScienceDirect and Google Scholar could be paid due attention to extend the coverage. The conclusive decision of both the databases shall reflect well the research field. Thus, the findings of the bibliometric and content analysis are limited to the Scopus database. Though some documents are indexed in more than one database, the inclusion of any of the databases could alter the findings of the bibliometric findings. Second, Even though we included most AI tools and techniques including ChatGPT and IPAs, not every tool was included in this study. keywords refinement can be done to identify more papers addressing other AI tools in the field. For instance, keywords identified through this study such as “Deep learning”, “machine learning”, “deep neural network”, “Machine translation” and “computer-aided instruction” could be included in future studies to extend the scope of the field. This study acknowledges the assumption that the identified keywords and content categories faithfully reflect the diversity and intricacies of AI tools employed in language learning. However, it is important to recognise that this approach may inadvertently overlook emerging trends or unconventional terminologies within this rapidly evolving field. Future research endeavors should remain attuned to these evolving nuances in the realm of AI tools for language learning. A different approach to review shall also be considered. This study has applied quantitative analysis to examine the research scope, similarly, studies could opt for qualitative analysis to extract valuable insights. Systematic reviews can be done on other prominent AI tools identified through content analysis. Given these limitations, the findings of the study can be beneficial for researchers in the field of AI in LL, since the study outlines both the research focus and the research gaps.

Ahmed S, Alshater MM, Ammari AE, Hammami H (2022) Artificial intelligence and machine learning in finance: a bibliometric review. Res Int Bus Financ 61:101646. https://doi.org/10.1016/j.ribaf.2022.101646

Article   Google Scholar  

Ali Z (2020) Artificial intelligence (AI): a review of its uses in language teaching and learning. IOP Conf Ser 769(1):012043. https://doi.org/10.1088/1757-899x/769/1/012043

Amaral L, Meurers D (2011) On using intelligent computer-assisted language learning in real-life foreign language teaching and learning. ReCALL 23(1):4–24. https://doi.org/10.1017/s0958344010000261

Bazeley KJP (2019) Qualitative data analysis with NVivo. CiNii Books

Belmonte JL, Moreno-Guerrero A-J, Núñez JAL, Lucena FJH (2020) Augmented reality in education. A scientific mapping in web of science. Interact Learn Environ 31(4):1860–1874. https://doi.org/10.1080/10494820.2020.1859546

Botero-Gomez V, Ruiz-Herrera LG, Arias AV, Díaz AR, Garnique JCV (2023) Use of virtual tools in teaching-learning processes: advancements and future direction. Soc Sci 12(2):70. https://doi.org/10.3390/socsci12020070

Briggs N (2018) Neural machine translation tools in the language learning classroom: students’ use, perceptions, and analyses. JALT CALL J 14(1):3–24. https://doi.org/10.29140/jaltcall.v14n1.221

Chen L, Chen P, Lin Z (2020) Artificial intelligence in education: a review. IEEE Access 8:75264–75278. https://doi.org/10.1109/access.2020.2988510

Chen X, Zou D, Cheng G, Xie H (2021) Artificial intelligence-assisted personalized language learning: systematic review and co-citation analysis. In: 2021 international conference on advanced learning technologies (ICALT), pp. 241–245 . IEEE

Chen X, Bear E, Hui B, Santhi-Ponnusamy H, Meurers D (2022) Education theories and AI affordances: Design and implementation of an intelligent computer assisted language learning system. In: Artificial intelligence in education. posters and late breaking results, workshops and tutorials, industry and innovation tracks, practitioners’ and doctoral consortium, pp 582–585. Springer

Chu H-C, Hwang G-H, Tu Y-F, Yang K-H (2022) Roles and research trends of artificial intelligence in higher education: a systematic review of the top 50 most-cited articles. Australas J Educ Technol 38(3):22–42

Google Scholar  

Cobo MJ, López-Herrera AG, Liu X, Herrera F (2011) An approach for detecting, quantifying, and visualizing the evolution of a research field: a practical application to the fuzzy sets theory field. J Inf 5(1):146–166. https://doi.org/10.1016/j.joi.2010.10.002

Cumming G, Sussex R, Cropp S (1993) Learning english as a second language: Towards the “mayday’’ intelligent educational system. Comput Educ. https://doi.org/10.1016/0360-1315(93)90078-w

Daǧdeler KO (2023) A systematic review of mobile-assisted vocabulary learning research. Smart Learn Environ. https://doi.org/10.1186/s40561-023-00235-z

Dizon G (2017) Using intelligent personal assistants for second language learning: a case study of Alexa. TESOL J 8(4):811–830. https://doi.org/10.1002/tesj.353

Dizon G (2021) Affordances and constraints of intelligent personal assistants for second-language learning. RELC J. https://doi.org/10.1177/00336882211020548

Dizon G, Tang D, Yamamoto Y (2022) A case study of using Alexa for out-of-class, self-directed Japanese language learning. Comput Educ Artif Intell 3:100088. https://doi.org/10.1016/j.caeai.2022.100088

Dogan ME, Dogan TG, Bozkurt A (2023) The use of artificial intelligence (AI) in online learning and distance education processes: a systematic review of empirical studies. Appl Sci 13(5):3056. https://doi.org/10.3390/app13053056

Dokukina I, Gumanova J (2020) The rise of chatbots-new personal assistants in foreign language learning. Proc Comput Sci 169:542–546. https://doi.org/10.1016/j.procs.2020.02.212

Donthu N, Kumar S, Mukherjee D, Pandey N, Lim WM (2021) How to conduct a bibliometric analysis: an overview and guidelines. J Bus Res 133:285–296. https://doi.org/10.1016/j.jbusres.2021.04.070

Esit O (2011) Your verbal zone: an intelligent computer-assisted language learning program in support of Turkish learners’ vocabulary learning. Comput Assist Lang Learn 24(3):211–232. https://doi.org/10.1080/09588221.2010.538702

Fang X, Ng DTK, Leung JKL, Chu S (2023) A systematic review of artificial intelligence technologies used for story writing. Educ Inf Technol. https://doi.org/10.1007/s10639-023-11741-5

Farrokhnia M, Banihashem SK, Noroozi O, Wals AEJ (2023) A SWOT analysis of ChatGPT: implications for educational practice and research. Innov Educ Teach Int. https://doi.org/10.1080/14703297.2023.2195846

Fryer LK, Ainley M, Thompson A, Gibson A, Sherlock Z (2017) Stimulating and sustaining interest in a language course: an experimental comparison of Chatbot and human task partners. Comput Hum Behav 75:461–468. https://doi.org/10.1016/j.chb.2017.05.045

Fryer LK, Nakao K, Thompson A (2019) Chatbot learning partners: connecting learning experiences, interest and competence. Comput Hum Behav 93:279–289. https://doi.org/10.1016/j.chb.2018.12.023

Gayed JM, Carlon MKJ, Oriola AM, Cross JS (2022) Exploring an AI-based writing assistant’s impact on english language learners. Comput Educ Artif Intell 3:100055. https://doi.org/10.1016/j.caeai.2022.100055

Godwin-Jones R (2022) Partnering with AI: intelligent writing assistance and instructed language learning. Lang Learn Technol 26(2):5–24. https://doi.org/10125/73474

Goodell JW, Kumar S, Lim WM, Pattnaik D (2021) Artificial intelligence and machine learning in finance: identifying foundations, themes, and research clusters from bibliometric analysis. J Behav Exp Financ 32:100577. https://doi.org/10.1016/j.jbef.2021.100577

Gruetzemacher R (2022) The power of natural language processing. https://hbr.org/2022/04/the-power-of-natural-language-processing

Gugerty L (2006) Newell and Simon’s logic theorist: historical background and impact on cognitive modeling. Proc Human Fact Ergon Soc Ann Meet 50(9):880–884. https://doi.org/10.1177/154193120605000904

Górriz JM, Ramírez J, Ortíz A, Martinez-Murcia FJ, Segovia F, Suckling J, Leming M, Zhang Y-D, Álvarez-Sánchez JR, Bologna G et al (2020) Artificial intelligence within the interplay between natural and artificial computation: advances in data science, trends and applications. Neurocomputing 410:237–270

Hariri W (2023) Unlocking the potential of chatgpt: A comprehensive exploration of its applications, advantages, limitations, and future directions in natural language processing. arXiv preprint arXiv:2304.02017

Home. https://hai.stanford.edu/

Hou Y, Yu Z (2023) A bibliometric analysis of synchronous computer-mediated communication in language learning using VOSViewer and CITNetExplorer. Educ Sci 13(2):125. https://doi.org/10.3390/educsci13020125

Huang W, Hew KF, Fryer LK (2021) Chatbots for language learning-Are they really useful? A systematic review of chatbot-supported language learning. J Comput Assist Learn 38(1):237–257. https://doi.org/10.1111/jcal.12610

Huang X, Zou D, Cheng G, Chen X, Xie H (2023) Trends, research issues and applications of artificial intelligence in language education. Educ Technol Soc 26(1):112–131

Jeon JH (2021) Chatbot-assisted dynamic assessment (CA-DA) for L2 vocabulary learning and diagnosis. Comput Assist Lang Learn. https://doi.org/10.1080/09588221.2021.1987272

Jeon JH, Lee S, Choi S (2023) A systematic review of research on speech-recognition chatbots for language learning: implications for future directions in the era of large language models. Interact Learn Environ. https://doi.org/10.1080/10494820.2023.2204343

Ji H, Han I, Ko Y (2022) A systematic review of conversational AI in language education: focusing on the collaboration with human teachers. J Res Technol Educ 55(1):48–63. https://doi.org/10.1080/15391523.2022.2142873

Katsarou E, Wild F, Sougari A-M, Chatzipanagiotou P (2023) A systematic review of voice-based intelligent virtual agents in EFL education. Int J Emerg Technol Learn (IJET) 18(10):65–85. https://doi.org/10.3991/ijet.v18i10.37723

Kessler G (2018) Technology and the future of language teaching. Foreign Lang Ann 51(1):205–218. https://doi.org/10.1111/flan.12318

Kienberger M, García-Holgado A, Schramm K, Raveling A, Meurers D, Labinska B, Koropatnitska T, Therón R (2022) Enhancing adaptive teaching of reading skills using digital technologies: the latill project. In: International conference on technological ecosystems for enhancing multiculturality, pp 1092–1098. Springer

Kohnke L, Moorhouse BL, Zou D (2023) ChatGPT for language teaching and learning. RELC J. https://doi.org/10.1177/00336882231162868

Krippendorff K (2018) Content analysis: an introduction to its methodology. Sage publications, California

Lee S, Jeon JH (2022) Visualizing a disembodied agent: young EFL learners’ perceptions of voice-controlled conversational agents as language partners. Comput Assist Lang Learn. https://doi.org/10.1080/09588221.2022.2067182

Liang J-C, Hwang G-J, Chen M-RA, Darmawansah D (2021) Roles and research foci of artificial intelligence in language education: an integrated bibliographic analysis and systematic review approach. Interact Learn Environ. https://doi.org/10.1080/10494820.2021.1958348

Lin MP-C, Chang D (2020) Enhancing post-secondary writers’ writing skills with a chatbot. J Educ Technol Soc 23(1):78–92

Litman D, Strik H, Lim GS (2018) Speech technologies and the assessment of second language speaking: approaches, challenges, and opportunities. Lang Assess Q 15(3):294–309. https://doi.org/10.1080/15434303.2018.1472265

Liu C, Liu S, Hwang G-J, Tu Y-F, Wang Y, Wang N (2023) Engaging EFL students’ critical thinking tendency and in-depth reflection in technology-based writing contexts: a peer assessment-incorporated automatic evaluation approach. Educ Inf Technol. https://doi.org/10.1007/s10639-023-11697-6

Liu Z, Wang S, Zhang Y, Feng Y, Li J, Zhu H (2023) Artificial intelligence in food safety: a decade review and bibliometric analysis. Foods 12(6):1242. https://doi.org/10.3390/foods12061242

Luo J (2022) A bibliometric review on artificial intelligence for smart buildings. Sustainability 14(16):10230. https://doi.org/10.3390/su141610230

Ma L (2021) An immersive context teaching method for college english based on artificial intelligence and machine learning in virtual reality technology. Mob Inf Syst 2021:1–7. https://doi.org/10.1155/2021/2637439

Meurers D (2012) Natural language processing and language learning. Encyc Appl Ling 4193–4205. https://www.sfs.uni-tuebingen.de/~dm/papers/meurers-13.pdf

Min W, Yu Z (2023) A bibliometric analysis of augmented reality in language learning. Sustainability 15(9):7235. https://doi.org/10.3390/su15097235

Nagata N (2013) An effective application of natural language processing in second language instruction. CALICO J 13(1):47–67. https://doi.org/10.1558/cj.v13i1.47-67

...Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann T, Mulrow CD, Shamseer L, Tetzlaff J, Akl EA, Brennan S, Chou R, Glanville J, Grimshaw J, Hróbjartsson A, Lalu MM, Li T, Loder E, Mayo-Wilson E, McDonald S, McGuinness LA, Stewart L, Thomas J, Tricco AC, Welch V, Whiting P, Moher D (2021) The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. https://doi.org/10.1136/bmj.n71

Peters ME, Ammar W, Bhagavatula C, Power R (2017) Semi-supervised sequence tagging with bidirectional language models. arXiv preprint arXiv:1705.00108

Pokrivčáková S (2019) Preparing teachers for the application of AI-powered technologies in foreign language education. J Lang Cult Educ 7(3):135–153. https://doi.org/10.2478/jolace-2019-0025

Popenici S, Kerr S (2017) Exploring the impact of artificial intelligence on teaching and learning in higher education. Res Pract Technol Enhanc Learn. https://doi.org/10.1186/s41039-017-0062-8

Rahman A, Tomy P (2023) Intelligent personal assistant–An interlocutor to mollify foreign language speaking anxiety. Interact Learn Environ. https://doi.org/10.1080/10494820.2023.2204324

Rahman A, Karthikeyan J, Tong C (2022) Appropriate Allocation of Specified NLP Tools for the Four Language Skills. Royal Book Publishing-International. https://doi.org/10.1080/10494820.2023.2204324

Ruiz S, Rebuschat P, Meurers D (2019) The effects of working memory and declarative memoryon instructed second language vocabulary learning: insights from intelligent CALL. Lang Teach Res 25(4):510–539. https://doi.org/10.1177/1362168819872859

Rusmiyanto R, Huriati N, Fitriani N, Tyas NK, Rofi’i A, Sari MN (2023) The role of artificial intelligence (AI) in developing english language learner’s communication skills. J Educ 6(1):750–757. https://doi.org/10.31004/joe.v6i1.2990

Russell S, Norvig P (2010) Artificial intelligence: a modern approach, 3rd edn. Prentice Hall, Hoboken

Segler TM, Pain H, Sorace A (2002) Second language vocabulary acquisition and learning strategies in ICALL environments. Comput Assist Lang Learn 15(4):409–422. https://doi.org/10.1076/call.15.4.409.8272

Seising R (2018) The emergence of fuzzy sets in the decade of the perceptron-Lotfi A. Zadeh’s and Frank Rosenblatt’s research work on pattern classification. Mathematics 6(7):110. https://doi.org/10.3390/math6070110

Shadiev R, Liu J (2022) Review of research on applications of speech recognition technology to assist language learning. ReCALL 35(1):74–88. https://doi.org/10.1017/s095834402200012x

Sharadgah TA, Sa’di RA (2022) A systematic review of research on the use of artificial intelligence in english language teaching and learning (2015–2021): what are the current effects? J Inf Technol Educ 21:337–377. https://doi.org/10.28945/4999

Shazly RE (2021) Effects of artificial intelligence on english speaking anxiety and speaking performance: a case study. Expert Syst. https://doi.org/10.1111/exsy.12667

Simonite T (2020) Baidu’s Artificial-Intelligence supercomputer beats Google at image recognition. MIT Technology Review. https://www.technologyreview.com/2015/05/13/168197/baidus-artificial-intelligence-supercomputer-beats-google-at-image-recognition/

Slavuj V, Kovačić B, Jugo I (2015) Intelligent tutoring systems for language learning. In: 38th international convention on information and communication technology, electronics and microelectronics (MIPRO). https://doi.org/10.1109/mipro.2015.7160383

Tai TY, Chen HH-J (2020) The impact of google assistant on adolescent EFL learners’ willingness to communicate. Interact Learn Environ. https://doi.org/10.1080/10494820.2020.1841801

Tan SC, Lee AVY, Lee M (2022) A systematic review of artificial intelligence techniques for collaborative learning over the past two decades. Comput Educ Artif Intell 3:100097. https://doi.org/10.1016/j.caeai.2022.100097

Thayyib PV, Mamilla R, Khan M, Fatima H, Asim M, Anwar I, Shamsudheen MK, Khan MA (2023) State-of-the-Art of artificial intelligence and big data analytics reviews in five different domains: a bibliometric summary. Sustainability 15(5):4026. https://doi.org/10.3390/su15054026

Triantafillou E, Pomportsis AS, Demetriadis S (2003) The design and the formative evaluation of an adaptive educational system based on cognitive styles. Comput Educ 41(1):87–103. https://doi.org/10.1016/s0360-1315(03)00031-9

Turing A (1950) I.-Computing machinery and intelligence. Mind LIX 236:433–460. https://doi.org/10.1093/mind/lix.236.433

Article   MathSciNet   Google Scholar  

Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN, Kaiser L, Polosukhin I (2017) Attention is all you need. Adv Neural Inf Process Syst. 30:5998–6008

Verma S, Gustafsson A (2020) Investigating the emerging COVID-19 research trends in the field of business and management: a bibliometric analysis approach. J Bus Res 118:253–261. https://doi.org/10.1016/j.jbusres.2020.06.057

Weitzman T (2023) GPT-4 released: what it means for the future of your business. Forbes. https://www.forbes.com/sites/forbesbusinesscouncil/2023/03/28/gpt-4-released-what-it-means-for-the-future-of-your-business/?sh=1567d4312dc6

Weng X, Chiu TKF (2023) Instructional design and learning outcomes of intelligent computer assisted language learning: systematic review in the field. Comput Educ Artif Intell 4:100117. https://doi.org/10.1016/j.caeai.2022.100117

Wijekumar K, Meyer BJF, Lei PW (2013) High-fidelity implementation of web-based intelligent tutoring system improves fourth and fifth graders content area reading comprehension. Comput Educ 68:366–379. https://doi.org/10.1016/j.compedu.2013.05.021

Xu Z, Wijekumar K, Ramirez G, Hu X, Irey R (2019) The effectiveness of intelligent tutoring systems on K-12 students’ reading comprehension: a meta-analysis. Br J Edu Technol 50(6):3119–3137. https://doi.org/10.1111/bjet.12758

Yang CT-Y, Lai SL, Chen HH-J (2022) The impact of intelligent personal assistants on learners’ autonomous learning of second language listening and speaking. Interact Learn Environ 1:21. https://doi.org/10.1080/10494820.2022.2141266

Yang H, Kyun S (2022) The current research trend of artificial intelligence in language learning: a systematic empirical literature review from an activity theory perspective. Australasian J Edu Technol 180–210. https://doi.org/10.14742/ajet.7492

Zhang R, Zou D, Cheng G (2023) A review of chatbot-assisted learning: pedagogical approaches, implementations, factors leading to effectiveness, theories, and future directions. Interact Learn Environ. https://doi.org/10.1080/10494820.2023.2202704

Ziegler N, Meurers D, Rebuschat P, Ruiz S, Moreno-Vega JL, Chinkina M, Li W, Grey S (2017) Interdisciplinary research at the intersection of CALL, NLP, and SLA: methodological implications from an input enhancement project. Lang Learn 67(S1):209–231. https://doi.org/10.1111/lang.12227

Zilio L, Wilkens R, Fairon C (2017) Using nlp for enhancing second language acquisition. In: RANLP, pp 839–846

Download references

The study received no external funding.

Author information

Antony Raj, Prajeesh Tomy and Mohamed Sahul Hameed have contributed equally to this work.

Authors and Affiliations

Department of English, School of Social Sciences and Languages, Vellore Institute of Technology, Vellore, Tamil Nadu, 632104, India

Abdur Rahman, Antony Raj, Prajeesh Tomy & Mohamed Sahul Hameed

You can also search for this author in PubMed   Google Scholar

Contributions

AR wrote the main manuscript made substantial contributions to the conception of the work. AR made substantial contributions to the conception, data analysis, and formatting of the work. PT drafted revised it critically for important intellectual content. MSH prepared drafted revised it critically for important intellectual content. All the authors reviewed the manuscript.

Corresponding author

Correspondence to Abdur Rahman .

Ethics declarations

Conflict of interest.

The authors of this paper declare no conflicts of interest related to the collection, analysis, interpretation, or presentation of the findings.

Ethical approval

We have adhered to rigorous ethical standards throughout the research process. We obtained all data from the Scopus database, and we acknowledge Scopus for providing access to this resource. We have complied with the terms and conditions of data usage set forth by Scopus.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Rahman, A., Raj, A., Tomy, P. et al. A comprehensive bibliometric and content analysis of artificial intelligence in language learning: tracing between the years 2017 and 2023. Artif Intell Rev 57 , 107 (2024). https://doi.org/10.1007/s10462-023-10643-9

Download citation

Accepted : 19 December 2023

Published : 01 April 2024

DOI : https://doi.org/10.1007/s10462-023-10643-9

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Artificial intelligence
• Language learning
• Bibliometric analysis
• Natural language processing
• Content analysis
• Find a journal
• Publish with us
• Track your research