phd thesis on cyberbullying

• Open access
• Published: 22 December 2021

Cyberbullying detection: advanced preprocessing techniques & deep learning architecture for Roman Urdu data

• Amirita Dewani   ORCID: orcid.org/0000-0002-3816-3644 1 ,
• Mohsin Ali Memon   ORCID: orcid.org/0000-0003-2638-4252 1 &
• Sania Bhatti   ORCID: orcid.org/0000-0002-0887-8083 1  

Journal of Big Data volume  8 , Article number:  160 ( 2021 ) Cite this article

10k Accesses

28 Citations

1 Altmetric

Metrics details

Social media have become a very viable medium for communication, collaboration, exchange of information, knowledge, and ideas. However, due to anonymity preservation, the incidents of hate speech and cyberbullying have been diversified across the globe. This intimidating problem has recently sought the attention of researchers and scholars worldwide and studies have been undertaken to formulate solution strategies for automatic detection of cyberaggression and hate speech, varying from machine learning models with vast features to more complex deep neural network models and different SN platforms. However, the existing research is directed towards mature languages and highlights a huge gap in newly embraced resource poor languages. One such language that has been recently adopted worldwide and more specifically by south Asian countries for communication on social media is Roman Urdu i-e Urdu language written using Roman scripting. To address this research gap, we have performed extensive preprocessing on Roman Urdu microtext. This typically involves formation of Roman Urdu slang- phrase dictionary and mapping slangs after tokenization. We have also eliminated cyberbullying domain specific stop words for dimensionality reduction of corpus. The unstructured data were further processed to handle encoded text formats and metadata/non-linguistic features. Furthermore, we performed extensive experiments by implementing RNN-LSTM, RNN-BiLSTM and CNN models varying epochs executions, model layers and tuning hyperparameters to analyze and uncover cyberbullying textual patterns in Roman Urdu. The efficiency and performance of models were evaluated using different metrics to present the comparative analysis. Results highlight that RNN-LSTM and RNN-BiLSTM performed best and achieved validation accuracy of 85.5 and 85% whereas F1 score was 0.7 and 0.67 respectively over aggression class.

Introduction

Cyberbullying (aka hate speech, cyberaggression and toxic speech) is a critical social problem plaguing today’s Internet users typically youth and lead to severe consequences like low self-esteem, anxiety, depression, hopelessness and in some cases causes lack of motivation to be alive, ultimately resulting in death of a victim [ 1 ]. Cyberbullying incidents can occur via various modalities. For example, it can take the form of sharing/ posting offensive video content or uploading violent images or sharing the pictures without permission of the owner etc. However, cyberbullying via textual content is far more common [ 2 ]. In Pakistan, the usage of internet, smartphones and social media has increasingly become prevalent these days and the very frequent users are youngsters. According to a report, more than 65% of all the users lie between 18 and 29, and typically women are more susceptible and unprotected. People often use offensive language, use hate speech, and become aggressive to bully celebrities, leaders, women and an individual [ 3 ]. In Pakistan, victims have reported life disturbing and annoying experiences and most of the victims are educated youngsters (age group of 21–30 years) [ 4 ]. The traffic in cyberspace has escalated significantly during covid-19 pandemic. A report “COVID 19 and Cyber Harassment”, released by DRF in 2020 highlights a great rise in the number of cyberbullying and harassment cases during the pandemic. The complaints registered with DRF’s Cyber Harassment Helpline were surged by 189% [ 5 ].

Recently, Roman Urdu language has been a contemporary trend and a viable language for communication on different social networking platforms. Urdu is national and official language of Pakistan and predominant among most communities across different regions. A survey statistic in [ 6 ] affirms that 300 million people are speaking Urdu language and approximately 11 million speakers are in Pakistan from which maximum users switched to Roman Urdu language for the textual communication, typically on social media. It is linguistically rich and morphologically complex language [ 7 ]. Roman Urdu language is highly variant with respect to word structures, writing styles, irregularities, and grammatical compositions. It is deficit of standard lexicon and available resources and hence become extremely challenging when performing NLP tasks.

An elaboration of script of Urdu instances and Roman Urdu is given in Table 1 . Instances highlighted are describing anti-social behavior.

This paper addresses toxicity/cyberbullying detection problem in Roman Urdu language using deep learning techniques and advanced preprocessing methods including usage of lexicons/resource that are typically developed to accomplish this work. Intricacies in analyzing the structure and patterns behind these typical aggressive behaviors, typically in a newly adopted language, and forming it as a comprehensive computational task is very complicated. The major contributions of this study are formation of a slang and contraction mapping procedure along with slang lexicon for Roman Urdu language and development of hybrid deep neural network models to capture complex aggression and bullying patterns.

The rest of the paper is organized as follows: Review of existing literature is presented in " Related Work " Section. " Problem statement " Section states research gap and gives formal definition of the addressed problem. " Methodology " Section describes the steps of research methodology and techniques and models used for the experimentations. Advanced preprocessing steps applied on Roman Urdu data are elaborated in " Data Preprocessing on Roman Urdu microtext " section. Implementation of proposed model architecture and hyperparameter settings are discussed in " Experimental Setup " section. " Results and Discussion " Section highlights and discusses study results and finally " Conlusion " Section concludes the research work and provides future research directions.

Related work

Due to the accretion of social media communication and adverse effects arising from its darker side on users, the field of automatic cyberbullying detection has become an emerging and evolving research trend [ 8 ]. Research work in [ 9 ] presents cyberbullying detection algorithm for textual data in English language. It is considered as one of the pioneers and highly cited research. They divided the task in text-classification sub problems related to sensitive topics and collected 4500 textual comments on controversial YouTube videos. This study implemented Naive Bayes, SVM and J48 binary and multiclass classifiers using general and specific feature sets. Study contributed in [ 10 ] applied deep learning architectures on Kaggle dataset and conducted experimental analysis to determine the effectiveness and performance of deep learning algorithms LSTM, BiLSTM, RNN and GRU in detecting antisocial behavior. Authors in [ 11 ] extracted data from four platforms i-e Twitter, YouTube, Wikipedia, and Reddit for developing an online hate classifier in English language using different classification techniques. Research carried out in [ 12 ] developed an automated approach to detect toxicity and unethical behavior in online communication using word embeddings and varying neural network layers. They suggested that LSTM layers and mimicked word embedding can uncover such behavior with good accuracy level.

Few of the studies in recent years has been contributed by researchers on other languages apart from English. Research work in [ 13 ] is unique and has gathered textual data from Instagram and twitter in Turkish language. They have implemented Naïve Bayes Multinomial, SVM, KNN and decision trees for cyberbullying detection along with Chi-square and information gain (IG) for feature selection. Work accomplished in [ 14 ] also addresses the problem of cyberaggression in Turkish language. The work extends comparison of different machine learning algorithms and found optimal results using Light Gradient Boosting Model. Van Hee, Cynthia, et al. in [ 15 ] proposed cyberbullying detection scheme for Dutch language. This is the first study on Dutch social media. Data was collected from ASKfm where users can ask and answer questions. The research uses default parameter settings for un-optimized linear kernel SVM based on n-grams and keyword system to identify bullying traces. F1 score for Dutch language was 61%. Problem of Arabic language cyberbullying detection was addressed and accomplished in [ 16 ]. This study used Dataiku DSS and WEKA for ML tasks. The data was scrapped from facebook and twitter. The study concluded that even though the detection approach was not comparable with the other studies in English language but overall Naive Bayes and SVM yield reasonable performance. Research work in [ 17 ] by Gomez-Adorno, Helena, et al. proposed automatic aggression detection for Spanish tweets. Several types of n-grams and linguistically motivated patterns were used but the best run could only achieve F1 score of 42.85%. Studies presented in [ 18 , 19 , 20 ] are based on automatic detection of cyberbullying content in German language. Research conducted in [ 18 ] proposed an approach based on SVM, CNN and ensemble model using unigram, bigrams and character N-grams for categorizing offensive tweets in German language. Research presented in [ 21 ] attempted for the very first time to identify bullying traces in Indonesian language. Association Rule mining and FP growth text mining were used to identify trends for bullying patterns in Jakarta and Surabaya cities using social media text. This baseline study on Indonesian language was further extended by Nurrahmi, Hani et al. in [ 22 ]. Study in [ 23 ] made first attempt to develop a corpus of code-mixed data considering Hindi and English language. They proposed a scheme for hate speech detection using N-grams and lexical features. An ensemble approach by combining the predictions of Convolutional Neural Network (CNN) and SVM algorithms were used for identifying such patterns. The weighted F1 score for Hindi language ranged between 0.37 and 0.55 for different experiments [ 24 ]. In the year 2019, Association for computational linguistics initiated the project for automatic detection of cyberbullying in Polish language [ 25 ]. Research conducted in [ 26 ] attempted to uncover cyberbullying patterns in Bengali language implementing passive aggressive, SVM and logistic regression. The optimum accuracy achieved was 78.1%. Recently, work contributed in [ 27 ] presented first study in Roman Urdu using lexicon based approach. The dataset was highly skewed comprising of only 2.2% toxic data. As according to [ 28 ], biased sampling and measurement errors are highly prone to classification errors when working on such datasets. Moreover, pattern detection based on predefined bullying and non-bullying lexicons were shortcomings of this study.

For automated detection of complex cyberbullying patterns, studies contributed by different scholars employ supervised, unsupervised, hybrid and deep learning models, vast feature engineering techniques, corpora, and social media platforms. However, the existing literature is mainly oriented towards unstructured data in English language. Some recent studies and projects have been initiated in other languages as discussed previously. To the best of our knowledge and literature review, no detailed work has been contributed in Roman Urdu to systematically analyze cyberbullying detection phenomenon using advanced preprocessing techniques (involving the usage of Roman Urdu resources) and deep learning approaches under different configurations.

Problem statement

The escalated usage of social networking sites and freedom of speech has given optimal ground to individuals across all demographics for cyberbullying and cyberaggression. This leaves drastic and noticeable impacts on behavior of a victim, ranging from disturbance in emotional wellbeing and isolation from society to more severe and deadly consequences [ 29 ]. Automatic Cyberbullying detection has remained very challenging task since social media content is in natural language and is usually posted in unstructured free-text form leaving behind the language norms, rules, and standards. Evidently, there exists a substantial number of research studies which primarily focus on discovering cyberbullying textual patterns over diverse social media platforms as discussed previously in literature review section. However, most of the detection schemes and automated approaches formulated are for resource-rich and mature languages spoken worldwide. Roman Urdu is typically spoken in South Asia and is a highly resource deficient language. Hence this research puts novel efforts to propose data pre-processing techniques on Roman Urdu scripting and develop deep learning-based hybrid models for automated cyberbullying detection in Roman Urdu language. The outcomes of this study, if implemented, will assist cybercrime centers and investigation agencies for monitoring social media contents and in making cyberspace secure and safer place for all segments of society.

Methodology

The research methodology is depicted in Fig.  1 .

Proposed research methodology

The development of hate speech/cyberbullying corpus with minor skew and automated development of domain specific roman Urdu stop words is published in our previous work [ 30 ]. The work details formation of computational linguist resources. Further steps of methodology are discussed in subsequent sections. The Deep Neural Network (DNN) based techniques and models used for the experimentations are detailed below.

Model description

Recurrent neural networks (rnn).

RNN [ 31 ] has been applied in literature for successive time series applications with temporal dependencies. An unfolded RNN can handle processing of current data by utilizing past data. Meanwhile, RNN has the issue of training long-term dependencies. This has been addressed by one of the RNN variant.

Long short-term memory networks (LSTM)

LSTM has been employed as an advanced version of RNN network. It resolves the shortcoming of RNN by applying memory cells also known as hidden layer units. Memory cells are controlled through three gates named as: input gate, output gate and forget gate. They have the self-connections which store the temporal state of network [ 31 ]. Input and output gates address and control the flow of information from memory cell input and output to rest of the network. The forget gate, usually called as a remember vector, transfers the information with higher weights from previous neuron to the next neuron. The forget gate is added to the memory cell. The information resides in memory depending upon the high activation results; the information will be stored in memory cell iff the input unit has high activation. However, the information will be transferred to next neuron if the output unit has high activation. Otherwise, input information with high weights resides in memory cell [ 31 ].

Mathematically, LSTM network can be described as [ 32 ]:

where W h   ∈  R m × d and U h   ∈  R m × m indicates weight matrices, x t denotes the current word embedding, b h   ∈  R m refers to bias term, whereas f(x ) is a non-linear function.

LSTM has more complex architecture including hidden states and tends to remember information for either short or long term. The hidden state [ 33 ] of LSTM is computed as follows:

where f t denotes the forget gate, i t refers to the input gate, c t denotes the cell state, o t is the output gate, h t is the regular hidden state, σ indicates sigmoid function, and ◦ is the Hadamard product.

Bidirectional Long short-term memory networks (BiLSTM)

In the traditional recurrent neural network model and LSTM model, the propagation of information is only in forward direction. This results in computation of an output vector only based on the current input at time t and the output of the previous unit. The back propagation of information in network is achieved by merging two bidirectional recurrent neural network (BiRNN) and LSTM units, one for forward direction and one for backward direction. This helps in capturing contextual information and enhances learning ability [ 34 ].

In bidirectional LSTM, outputs of two LSTM networks are stacked together. The first LSTM is a regular sequence starting from the starting of the paragraph, while the second LSTM is a standard sequence, and the series of inputs are fed in the opposite order. The first hidden state is denoted by ht forward whereas second LSTM unit’s hidden state is denoted by ht backward . After processing data, the final state ht Bilstm is computed by concatenating the two hidden states as given in Eq.  3 .

where  ⊕  denotes a concatenation operator.

Convolutional neural networks (CNN)

Convolution neural networks (aka CNN), originally incorporated for image processing tasks, have become very efficacious in different NLP and text classification applications. The network identifies correlations and patterns of data via their feature maps. Information about local structure of data is extracted by applying multiple filters with different dimensions [ 35 ].

Data preprocessing on Roman Urdu microtext

Big social media data in Roman Urdu language is inconsistent, incomplete, or precise, missing in certain behaviors or trends, and is likely to incorporate many errors. Roman Urdu users highly deviate language norms while communicating on social media. Hence data preprocessing is immensely significant. Some major data preprocessing steps applied on Roman Urdu microtext are detailed below.

Handling Unicode and encoded text formats

Unicode scheme provides every character in natural language text a unique code from 0 to 0 × 10FFFF. The uncleaned Roman Urdu data comprised of special symbols, emojis, and other typical stray characters represented using Unicode. We used Unicode transformation type 8 encoding to convert the data. This data was converted and handled using re and string modules in python.

Text cleaning

Text cleaning is essential step to eliminate or at least reduce noise from Microtext. This step comprised of case transformation, removal of punctuations and URLs, elimination of additional white spaces, exclusion of hashtags, digits & special character removal and removal of metadata/non-linguistic features.

Tokenization

Tokenization is immensely essential phase of text processing. It is the process of generating tokens by splitting textual content into words, phrases, or other meaningful parts. It is generally a form of text segmentation [ 36 ]. Tokenization was performed using Keras tokenizer to prepare the text for implementing deep learning networks.

Filtering stop words

Stop words are non-semantic division of text in natural language. The necessity that stop words should be eliminated from text is that they make the text higher dimensional with redundant features which are less significant for analysts. Removing stop words reduces the dimensionality of term space [ 37 ]. Development of domain specific stop words in Roman Urdu language automatically using statistical techniques and bilingual experts’ input, comprising of 173 words is detailed in our previous work [ 30 ]. Insignificant Roman Urdu words were typically articles such as ek (ایک), conjunctions and pronouns such as tum (تم), tumhara (تمھارا), us (اُس), wo/who (وہ), usko (اُسکو), preposition such as main (میں), pe (پے), par (پر), demonstratives such as ye (یے), inko (انکو), yahan(یہاں), and interrogatives such as kahan (کہاں), kab (کب), kisko (کس کو), kiski (کس کی) etc. Stop words were removed from Roman Urdu corpus, leaving behind the index terms which are important.

Mapping slangs and contractions

Existing libraries, APIs and toolkits in python language primarily support preprocessing functions for English and other mature languages. They can be partially used for Roman Urdu language. Moreover, most of the communication in Roman Urdu comprises of bully terms being used as slangs. High dimensional textual data also suppress significant features. Hence contraction mapping is mandatory for dimensionality reduction and to capture complex bullying patterns. Currently, Pycontractions Library [ 38 ] only supports English contraction mapping process. To address this problem, the study developed data slang mapping process. To map slangs to original terms and phrases in Roman Urdu language, we created Slang lexicon in Roman Urdu (SLRU) which also included Roman Urdu abuses and offensive terms used as a norm by Roman Urdu users. SLRU is in the form of a dictionary. It comprises of the key: value pairs, where key is the slang and value is its equivalent Roman Urdu phrase/term such as “AFIK”: “Jahan tak mujhay pata hai”, ASAP: “Jitna jaldi ho sakay”, “tbh”: “Sach main” and so on. The process of slang mapping is detailed in Fig.  2 .

Mapping process for slangs in Roman Urdu

The results of mapping process are highlighted in Fig.  3 .

Mapping on Roman Urdu Data

Experimental setup

This section discusses implementation of proposed neural network architecture and all hyper-parameter settings. All the experiments were performed on 11 Gen, core i7, 4 cores, 8 logical microprocessors, with 2.8 GHz processor speed, 256 GB Solid State Drive and python version 3.8, 64 bits.

Proposed model implementation and hyper-parameter settings

All models were implemented and trained in Keras; a high-level neural network API that works with open-source machine learning framework called TensorFlow [ 39 ]. All the implementation was accomplished using PyCharm. The optimal parameters and results were achieved through repeated experimentations.

Data was split into training and testing datasets. The data split was 0.8 for training and 0.2 for testing i-e 80% of data instances were used for training and 20% were holdout for testing and validation purpose. The sets were made using shuffled array. This allows model to learn over different data instances. Moreover, it helps to uncover reliability of model and consistency of results over repeated executions. Random state is also generated using numpy.random [ 40 ] for random sampling during splitting of data to ensure reproducible splits.

Textual input data must be integer encoded. In RNN-LSTM architecture, a sequential model was created. Initially an Embedding layer was added to the network and textual Roman Urdu data was provided as an input. Embedding layer embedded high dimensional text data in low dimensional vector space for generating dense vector representation of data. Embedding was formed using 2000 features and 128 embed dimensions. The experiment was initially executed on 20 epochs and 50 batch size. The batch size was based on the fact that model was having single lstm layer, and comparatively took lower training and validation time per step. The execution time for each epoch was approximately 10 ms. SpatialDropout1D was used with rate of 0.3. It helped to regularize the activations and maintain effective learning rate of the model. For updating network weights iteratively, this work uses binary cross entropy loss function and Adam optimizer. Sigmoid activation function was also implemented. It is denoted by f(x) and is defined as:

The Spatial Dropout layer was implemented instead of a simple Dropout. The major reason being was to retain the context of textual data established by neighboring words. Dropping random words (except for stop words, which were already removed during preprocessing step) can highly affect the context of uttered sentences and ultimately the performance of model. We incorporated two hidden dense layers denoted by D 1 and D 2 . The output of each hidden layer was computed to get the final output for cyberbullying text detection.

Keras tokenizer was used to accomplish pre-tokenization of all the data required for the implementation of RNN-biLSTM model. We created a sequential model with Embedding layer having 2000 maximum features. Subsequently a biLSTM layer comprising of two LSTM layers, one to read sequence in forward direction and other in backward direction, each with 64 units was added. Hidden layer (H 1 ) was formed using sigmoid activation function. For down sampling the feature maps, Dropout layer was added with 0.2 dropout rate. Moreover, we used 128 batch size to utilize low to moderate computational resources while still not slowing down the training process. Batch size highlights number of samples processed by model before updating of internal parameters. To combat overfitting, we added second dropout layer with rate of 0.25. Adam optimization was used with learning rate of 0.01 since batch size was not too small. For this model, we used binary cross entropy loss function. As the Epochs increase, the generalization ability of the model improves. However, too many epochs also lead to the problem of overfitting. The model was executed over different number of Epochs and average execution time for each epoch was 13 to 15 ms. The performance of model stabilized over 20 Epochs, above which the improvement was almost negligible.

In CNN model, initially the sentence was transformed into matrix where each row of matrix represented word vectors representation of data. We used 1000 features and 32 dimensions. Two convolutional filters were applied with 8 and 16 filters and 3 kernel size. Each filter was used to perform one dimensional convolution on word embeddings. Both Layers were 1D in nature. We set two dropout layers with dropout rate of 0.25 to improve generalization ability of developed model. Hidden layers with Relu and sigmoid activation functions were used. To extract most salient and prominent features, global maximum pooling layers were used with pool size = 2. Flatten layer was created after convolutional layers to flatten the output of the previous layer to a single long feature vector. The experiment was simulated over different Epochs. However, results got stable at 30 epochs.

Results and discussion

Empirical evaluation of cyberbullying detection scheme performance in Roman Urdu and experimental setups is accomplished via accuracy, precision, recall, and f1 measure metrics.

All the implemented models were executed several times over number of epochs to get consistency in evaluation parameters until it was a minor difference of ± 0.1. The results for LSTM are depicted in Fig.  4 . To ensure results validity and reliability, for a comparatively less skewed dataset, F1 measure (i-e a harmonic mean of precision and recall) is used as an evaluation metric. Furthermore, we have also reported macro and weighted average scores across all the classes. The evaluation results of RNN-LSTM are given in Fig.  4 .

RNN-LSTM evaluation Results

F1 score for RNN-LSTM over cyberbullying class was only 70%, however for non-cyberbullying class, score was 90%. We observed that nearly all the instances of majority class of non-cyber bullying are correctly classified by this model. The experimental simulation depicting model accuracy and validation accuracy during training and validation phases, before and after stabilization of evaluation parameters is represented in Figs.  5 and 6 respectively.

RNN-LSTM Model accuracy graph for 20 epochs

RNN-LSTM Model accuracy graph for 50 epochs

The accuracy improved over subsequent epochs. However, after 20 epochs it got stabilized. The average accuracy produced by this model was 93.5% during training and 85.5% during validation. Overall curve variation is indicating that no overfitting problem arise. The model loss during training and validation loss during testing over 20 and 50 epochs is shown in Figs.  7 and 8 respectively. The cross-entropy loss considered during configuration over different epochs converged well, thus indicating optimal model performance.

RNN-LSTM Model loss plot- Binary Cross entropy for 20 epochs

RNN-LSTM Model loss plot- Binary Cross entropy for 50 epochs

The evaluation results of RNN-BILSTM model over 20 epochs are given in Fig.  9 .

RNN-BiLSTM evaluation Results

RNN-LSTM also performed reasonably well for cyberbullying detection task on Roman Urdu data. F1 score for non-cyberbullying content prediction was 90% whereas for cyberbullying content, the score was 67% only. This indicates that model erroneously classified/misclassified some of the aggressive class instances and TN rate was at average. Figs.  10 and 11 are depicting model accuracy and validation accuracy for RNN-BiLSTM.

RNN-biLSTM Model accuracy plot for 20 epochs

RNN-biLSTM Model accuracy plot for 50 epochs

The accuracy improved highly during training process up to 20 epochs. Overall average accuracy was 97% in training and 85% on validation set. 20% of the data was used for as a validation set, as stated earlier. During experimentation, we identified that accuracy of our model is not improving after a specific point i-e after 20 Epochs. The trivial variations can be clearly visualized from the graph in Fig.  9 . Model loss and validation loss during training and testing process for RNN-BiLSTM over 20 and 50 epochs is given in Figs.  12 and 13 respectively.

RNN-BiLSTM Model loss plot- Binary Cross entropy for 20 epochs

RNN-BiLSTM Model loss plot- Binary Cross entropy for 50 epochs

The cross-entropy loss was minimal (approximately 1.2), indicating good prediction capability of developed model.

Figure  14 represents the evaluation results for CNN model.

CNN model evaluation Results

CNN performed well for prediction of non-cyberbullying content, providing F1 score of 87%. However, model did not yield good efficiency for categorizing cyberbullying class, producing f1-score of 52%. The repeated experiments performed for CNN showed continuous improvements up to 30 Epochs. Figure  15 depicts model accuracy and validation accuracy. The experimental simulation over 50 epochs only shown minor improvements as represented in Fig.  16 . The average execution time for Epoch was 9 ms each. The training accuracy of 98% was achieved over different executions whereas model produced 85% validation accuracy.

CNN Model accuracy plot for 30 epochs

CNN Model accuracy plot for 50 epochs

CNN model loss and validation loss results at 30 and 50 epochs are presented in Fig.  17 and 18 respectively. The loss was minimal during training and converged. During validation the loss increased and diverged indicating only moderate performance over unseen instances typically from aggressive class.

CNN Model loss plot- Binary Cross entropy for 30 epochs

CNN Model loss plot- Binary Cross entropy for 50 epochs

The compiled model results indicating evaluation measures at stabilized epochs are depicted in Table 2 .

Cyberbullying has become an alarming social threat for today’s youth and has recently gained huge attention from research community. This research has addressed the problem of cyberbullying detection in Roman Urdu Language. Since Roman Urdu is highly resource deficient language, having different writing patterns, word structures, and irregularities thus making this work a challenging task. In this work we have presented advanced preprocessing techniques mainly a slang mapping mechanism, domain specific stop word removal, handling encoded formats and formulation of deep learning architecture to detect cyberbullying patterns in Roman Urdu language. We created experiments with vast parameters to build optimal classifier for cyberbullying tweets. The results highlighted that RNN-LSTM and RNN-BiLSTM with concatenation of forward and backward units provided better performance in 20 Epochs as compared to CNN. The existing work can be extended in numerous ways. The future studies can focus on development of ensemble models to uncover harassing and hate speech patterns. Moreover, the incorporation of context-specific features and handling of morphological variations might produce better results.

Availability of data and materials

The used raw dataset in this research is not publicly available. The data that support the findings of this research work are available from the corresponding author, on valid request due to privacy and ethical restrictions.

Abbreviations

Recurrent neural network

Long-short term memory

Bidirectional long-short term memory

Convolutional neural network

Social networking

Right to left

Left to right

True negative

Hellfeldt K, López-Romero L, Andershed H. Cyberbullying and psychological well-being in young adolescence: the potential protective mediation effects of social support from family, friends, and teachers. Int J Environ Res Public Health. 2020;17(1):45.

Article   Google Scholar  

Dadvar M. Experts and machines united against cyberbullying [PhD thesis]. University of Twente. 2014.

Magsi H, Agha N, Magsi I. Understanding cyber bullying in Pakistani context: causes and effects on young female university students in Sindh province. New Horiz. 2017;11(1):103.

Google Scholar  

Qureshi SF, Abbasi M, Shahzad M. Cyber harassment and women of Pakistan: analysis of female victimization. J Bus Soc Rev Emerg Econ. 2020;6(2):503–10.

S. Irfan Ahmed, Cyber bullying doubles during pandemic. https://www.thenews.com.pk/tns/detail/671918-cyber-bullying-doubles-during-pandemic . Accessed 24 Aug 2020.

Shahroz M, Mushtaq MF, Mehmood A, Ullah S, Choi GS. RUTUT: roman Urdu to Urdu translator based on character substitution rules and unicode mapping. IEEE Access. 2020;8:189823–41.

Mehmood F, Ghani MU, Ibrahim MA, Shahzadi R, Mahmood W, Asim MN. A precisely xtreme-multi channel hybrid approach for roman urdu sentiment analysis. IEEE Access. 2020;8:192740–59.

Alotaibi M, Alotaibi B, Razaque A. A multichannel deep learning framework for cyberbullying detection on social media. Electronics. 2021;10(21):2664.

Dinakar K, Reichart R, Lieberman H. Modeling the detection of textual cyberbullying. In: 5th international AAAI conference on weblogs and social media. 2011.

Iwendi C, Srivastava G, Khan S, Maddikunta PKR. Cyberbullying detection solutions based on deep learning architectures. Multimed Syst. 2020. https://doi.org/10.1007/s00530-020-00701-5 .

Salminen J, Hopf M, Chowdhury SA, Jung S, Almerekhi H, Jansen BJ. Developing an online hate classifier for multiple social media platforms. Hum Cent Comput Inf Sci. 2020;10(1):1–34.

Dessì D, Recupero DR, Sack H. An assessment of deep learning models and word embeddings for toxicity detection within online textual comments. Electronics. 2021;10(7):779.

S. A. Özel, E. Saraç, S. Akdemir, and H. Aksu, Detection of cyberbullying on social media messages in Turkish, In: 2017 International Conference on Computer Science and Engineering (UBMK), 2017, pp. 366–370.

E. C. Ates, E. Bostanci, and M. S. Guzel, Comparative Performance of Machine Learning Algorithms in Cyberbullying Detection: Using Turkish Language Preprocessing Techniques, arXiv Prepr. arXiv2101.12718, 2021.

Van Hee C, et al. Automatic detection of cyberbullying in social media text. PLoS ONE. 2018;13(10):e0203794.

Haidar B, Chamoun M, Serhrouchni A. A multilingual system for cyberbullying detection: Arabic content detection using machine learning. Adv Sci Technol Eng Syst J. 2017;2(6):275–84.

Gómez-Adorno H, Bel-Enguix G, Sierra G, Sánchez O, Quezada D. A machine learning approach for detecting aggressive tweets in Spanish, In: IberEval@ SEPLN. 2018. pp. 102–107.

X. Bai, F. Merenda, C. Zaghi, T. Caselli, and M. Nissim, RuG at GermEval: Detecting Offensive Speech in German Social Media, in 14th Conference on Natural Language Processing KONVENS 2018, 2018, p. 63.

B. Birkeneder, J. Mitrovic, J. Niemeier, L. Teubert, and S. Handschuh, upInf—Offensive Language Detection in German Tweets, In: Proceedings of the GermEval 2018 Workshop, 2018, pp. 71–78.

J. M. Schneider, R. Roller, P. Bourgonje, S. Hegele, and G. Rehm, Towards the Automatic Classification of Offensive Language and Related Phenomena in German Tweets, In: 14th Conference on Natural Language Processing KONVENS 2018, 2018, p. 95.

H. Margono, X. Yi, and G. K. Raikundalia, Mining Indonesian cyber bullying patterns in social networks, In: Proceedings of the Thirty-Seventh Australasian Computer Science Conference-Volume 147, 2014, pp. 115–124.

H. Nurrahmi and D. Nurjanah, Indonesian Twitter Cyberbullying Detection using Text Classification and User Credibility, In: 2018 International Conference on Information and Communications Technology (ICOIACT), 2018, pp. 543–548.

A. Bohra, D. Vijay, V. Singh, S. S. Akhtar, and M. Shrivastava, A Dataset of Hindi-English Code-Mixed Social Media Text for Hate Speech Detection, In: Proceedings of the Second Workshop on Computational Modeling of People’s Opinions, Personality, and Emotions in Social Media, 2018, pp. 36–41.

A. Roy, P. Kapil, K. Basak, and A. Ekbal, An ensemble approach for aggression identification in english and hindi text, In: Proceedings of the First Workshop on Trolling, Aggression and Cyberbullying (TRAC-2018), 2018, pp. 66–73.

Association for Computational Linguistics. https://www.aclweb.org/portal/content/deadline-extension-first-task-automatic-cyberbullying-detection-polish-language . Accessed 09 May 2019.

Ghosh R, Nowal S, Manju G. Social media cyberbullying detection using machine learning in bengali language. Int J Eng Res Technol. 2021. https://doi.org/10.1109/ICECE.2018.8636797 .

Talpur KR, Yuhaniz SS, Sjarif NNBA, Ali B. Cyberbullying detection in Roman Urdu language using lexicon based approach. J Crit Rev. 2020;7(16):834–48. https://doi.org/10.31838/jcr.07.16.109 .

J. Brownlee, Imbalanced Classification, December 23, 2019. https://machinelearningmastery.com . Accessed 10 May 2021.

Arif M. A systematic review of machine learning algorithms in cyberbullying detection: future directions and challenges. J Inf Secur Cybercrimes Res. 2021;4(1):1–26.

A. Dewani, M. Ali Memon, and S. Bhatti, Development of Computational Linguistic Resources for Automated Detection of Textual Cyberbullying Threats in Roman Urdu Language, 3C TIC. Cuad. Desarro. Apl. a las TIC, 101–121., p. 17, 2021.

Shahid F, Zameer A, Muneeb M. Predictions for COVID-19 with deep learning models of LSTM, GRU and Bi-LSTM. Chaos Solitons Fractals. 2020;140:110212.

M. Cliche, “BB_twtr at SemEval-2017 task 4: Twitter sentiment analysis with CNNs and LSTMs,” arXiv Prepr. arXiv1704.06125, 2017.

W. Zaremba, I. Sutskever, and O. Vinyals, Recurrent neural network regularization, arXiv Prepr. arXiv1409.2329, 2014.

Xu G, Meng Y, Qiu X, Yu Z, Wu X. Sentiment analysis of comment texts based on BiLSTM. Ieee Access. 2019;7:51522–32.

S. Minaee, E. Azimi, and A. Abdolrashidi, Deep-sentiment: Sentiment analysis using ensemble of cnn and bi-lstm models, arXiv Prepr. arXiv1904.04206, 2019.

Uysal AK, Gunal S. The impact of preprocessing on text classification. Inf Process Manag. 2014;50(1):104–12.

Vijayarani S, Ilamathi MJ, Nithya M. Preprocessing techniques for text mining-an overview. Int J Comput Sci Commun Networks. 2015;5(1):7–16.

Pycontractions 2.0.1. https://pypi.org/project/pycontractions/ . Accessed 17 Nov 2021.

API Documentation. https://www.tensorflow.org/api_docs . Accessed 21 Oct 2020.

NumPy. https://numpy.org/ . Accessed 18 Nov 2021.

Download references

Acknowledgements

We would like to thank Institute of Information and Communication Technology, Mehran University of Engineering & Technology, for providing resources and funding, necessary to accomplish this research work.

This research has been performed at Institute of Information and Communication Technology, Mehran University of Engineering and Technology, Pakistan and is fully funded under MUET funds for postgraduate students.

Author information

Authors and affiliations.

Institute of Information and Communication Technologies, Department of Software Engineering, Mehran University of Engineering & Technology, Jamshoro, Sindh, Pakistan

Amirita Dewani, Mohsin Ali Memon & Sania Bhatti

You can also search for this author in PubMed   Google Scholar

Contributions

Corresponding author is the main contributor of this work from conception, drafting, algorithms development and implementation to analysis of results. Other authors are research supervisors who provided valuable guidance for work and content improvement. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Amirita Dewani .

Ethics declarations

Ethics approval and consent to participate.

Not applicable.

Consent for publication

Competing interests.

The authors declare that they have no competing or conflicts of interest to report regarding the present study.

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

Cite this article.

Dewani, A., Memon, M.A. & Bhatti, S. Cyberbullying detection: advanced preprocessing techniques & deep learning architecture for Roman Urdu data. J Big Data 8 , 160 (2021). https://doi.org/10.1186/s40537-021-00550-7

Download citation

Received : 03 September 2021

Accepted : 10 December 2021

Published : 22 December 2021

DOI : https://doi.org/10.1186/s40537-021-00550-7

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

• Advanced preprocessing
• Deep learning
• Hate speech detection
• Cyberbullying

Experts and Machines United Against Cyberbullying

Research output : Thesis › PhD Thesis - Research UT, graduation UT

• METIS-304975
• Information Retrieval
• Sentiment Analysis
• Cyberbullying
• Expert Systems

Access to Document

• 10.3990/1.9789036537391
• thesis_M_Dadvar Final published version, 1.25 MB
• DISSERTATION_84_eprint Final published version, 2.22 MB

T1 - Experts and Machines United Against Cyberbullying

AU - Dadvar, M.

N1 - SIKS Dissertation series no. 2014-37

PY - 2014/9/12

Y1 - 2014/9/12

N2 - One form of online misbehaviour which has deeply affected society with harmful consequences is known as cyberbullying. Cyberbullying can simply be defined as an intentional act that is conducted through digital technology to hurt someone. Cyberbullying is a widely covered topic in the social sciences. There are many studies in which the problem of cyberbullying has been introduced and its origins and consequences have been explored in detail. There are also studies which have investigated the intervention and prevention strategies and have proposed guidelines for parents and adults in this regard. However, studies on the technical dimensions of this topic are relatively rare. In this research the overall goal was to bridge the gap between social science approaches and technical solutions. In order to be able to suggest solutions that could contribute to minimizing the risk and impact of cyberbullying we have investigated the phenomenon of cyberbullying from different angles. We have thoroughly studied the origin of cyberbullying and its growth over time, as well as the role of technology in the emergence of this type of virtual behaviour and in the potential for reducing the extent of the social concern it raises. First we introduced a novel outlook towards the cyberbullying phenomenon. We looked into the gradual changes which have occurred in relationships and social communication with the emergence of the Internet. We argued that one should look at virtual environments as virtual communities, because the human needs projected on these environments, the relationships, human concerns and misbehaviour have the same nature as in real-life societies. Therefore, to make virtual communities safe, we need to take safety measures and precautions that are similar to the ones that are common in non-virtual communities. We derived the assumption that if cyberbullying is recognized and treated as a social problem and not just seen as some random mischief conducted by individuals with the use of technology, the methods for handling its consequences are likely to be more realistic, effective and comprehensive. This part of our study led to the conviction that for combating cyberbullying, behavioural and psychological studies, and the study of technical solutions should go hand in hand. One of the main limitations that we faced when we started our research was the lack of a comprehensive dataset for cyberbullying studies. We needed a dataset which included real instances of bullying incidents. Moreover, it was essential for our studies to also have the demographic information of the social media users as well as the history of their activities. We started our preliminary experiments using a dataset that was collected from MySpace forums. This dataset did not meet all the requirements for our experiment, namely in terms of size and sufficiency of information. Therefore we developed our own YouTube dataset, with the aim to encompass extensive information about the users and their activities as well as larger numbers of bullying comments. We collected information on user activities and posted textual comments as well as personal and demographic details of the users involved. Detecting a bullying comment or post at the earliest possible moment in time can substantially decrease the negative effects of cyberbullying incidents. We started our experiments by showing that besides the conventional features used for text mining methods such as sentiment analysis and specifically bullying detection, more personal features, in this experiment gender, can improve the accuracy of the detection models. As expected the models which were optimized accordingly resulted in a more accurate classification. The improved outcome motivated us to look into other personal features as well, such as age and the writing style of users. By adding more personal information, the previous classification results were outperformed and the detection accuracy enhanced even further. In the last experiment we made use of experts’ knowledge to identify potential bully users in social networks. To better understand and interpret the intentions underlying the online activities of users of social media, we decided to incorporate human reasoning and knowledge into a bulliness rating system by developing a Multi-Criteria Evaluation System. Moreover, to have more sources of information and to make use of the potential of both human and machine, we designed a hybrid approach, incorporating machine learning models on top of the expert system. The hybrid approach reached an optimum model which outperformed the results obtained from the machine learning models and the expert system individually. Our hybrid model illustrates the added value of integrating technical solutions with insights from the social sciences for the first time. As argued in this thesis, the integration of social studies into a software-enhanced monitoring workflow could pave the way towards the tackling of this kind of online misbehaviour. The ideas and algorithms proposed for fulfilling this purpose can be a stepping stone for future research in this direction. The work carried out is also a demonstration of the added value of frameworks for text categorization, sentiment mining and user profiling in applications addressing societal issues. This work can be viewed as a contribution to the more general societal challenge of increasing the level of cybersecurity, in particular for the younger generations of social network users. By turning the internet into a safer place for children, the chances increase that they will be able to benefit from the informational richness that it also offers.

AB - One form of online misbehaviour which has deeply affected society with harmful consequences is known as cyberbullying. Cyberbullying can simply be defined as an intentional act that is conducted through digital technology to hurt someone. Cyberbullying is a widely covered topic in the social sciences. There are many studies in which the problem of cyberbullying has been introduced and its origins and consequences have been explored in detail. There are also studies which have investigated the intervention and prevention strategies and have proposed guidelines for parents and adults in this regard. However, studies on the technical dimensions of this topic are relatively rare. In this research the overall goal was to bridge the gap between social science approaches and technical solutions. In order to be able to suggest solutions that could contribute to minimizing the risk and impact of cyberbullying we have investigated the phenomenon of cyberbullying from different angles. We have thoroughly studied the origin of cyberbullying and its growth over time, as well as the role of technology in the emergence of this type of virtual behaviour and in the potential for reducing the extent of the social concern it raises. First we introduced a novel outlook towards the cyberbullying phenomenon. We looked into the gradual changes which have occurred in relationships and social communication with the emergence of the Internet. We argued that one should look at virtual environments as virtual communities, because the human needs projected on these environments, the relationships, human concerns and misbehaviour have the same nature as in real-life societies. Therefore, to make virtual communities safe, we need to take safety measures and precautions that are similar to the ones that are common in non-virtual communities. We derived the assumption that if cyberbullying is recognized and treated as a social problem and not just seen as some random mischief conducted by individuals with the use of technology, the methods for handling its consequences are likely to be more realistic, effective and comprehensive. This part of our study led to the conviction that for combating cyberbullying, behavioural and psychological studies, and the study of technical solutions should go hand in hand. One of the main limitations that we faced when we started our research was the lack of a comprehensive dataset for cyberbullying studies. We needed a dataset which included real instances of bullying incidents. Moreover, it was essential for our studies to also have the demographic information of the social media users as well as the history of their activities. We started our preliminary experiments using a dataset that was collected from MySpace forums. This dataset did not meet all the requirements for our experiment, namely in terms of size and sufficiency of information. Therefore we developed our own YouTube dataset, with the aim to encompass extensive information about the users and their activities as well as larger numbers of bullying comments. We collected information on user activities and posted textual comments as well as personal and demographic details of the users involved. Detecting a bullying comment or post at the earliest possible moment in time can substantially decrease the negative effects of cyberbullying incidents. We started our experiments by showing that besides the conventional features used for text mining methods such as sentiment analysis and specifically bullying detection, more personal features, in this experiment gender, can improve the accuracy of the detection models. As expected the models which were optimized accordingly resulted in a more accurate classification. The improved outcome motivated us to look into other personal features as well, such as age and the writing style of users. By adding more personal information, the previous classification results were outperformed and the detection accuracy enhanced even further. In the last experiment we made use of experts’ knowledge to identify potential bully users in social networks. To better understand and interpret the intentions underlying the online activities of users of social media, we decided to incorporate human reasoning and knowledge into a bulliness rating system by developing a Multi-Criteria Evaluation System. Moreover, to have more sources of information and to make use of the potential of both human and machine, we designed a hybrid approach, incorporating machine learning models on top of the expert system. The hybrid approach reached an optimum model which outperformed the results obtained from the machine learning models and the expert system individually. Our hybrid model illustrates the added value of integrating technical solutions with insights from the social sciences for the first time. As argued in this thesis, the integration of social studies into a software-enhanced monitoring workflow could pave the way towards the tackling of this kind of online misbehaviour. The ideas and algorithms proposed for fulfilling this purpose can be a stepping stone for future research in this direction. The work carried out is also a demonstration of the added value of frameworks for text categorization, sentiment mining and user profiling in applications addressing societal issues. This work can be viewed as a contribution to the more general societal challenge of increasing the level of cybersecurity, in particular for the younger generations of social network users. By turning the internet into a safer place for children, the chances increase that they will be able to benefit from the informational richness that it also offers.

KW - METIS-304975

KW - Information Retrieval

KW - EWI-25014

KW - Sentiment Analysis

KW - Cyberbullying

KW - Expert Systems

KW - IR-91720

U2 - 10.3990/1.9789036537391

DO - 10.3990/1.9789036537391

M3 - PhD Thesis - Research UT, graduation UT

SN - 978-90-365-3739-1

PB - University of Twente

CY - Enschede

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
• Front Public Health

Cyberbullying Among Adolescents and Children: A Comprehensive Review of the Global Situation, Risk Factors, and Preventive Measures

Chengyan zhu.

1 School of Political Science and Public Administration, Wuhan University, Wuhan, China

Shiqing Huang

2 School of Medicine and Health Management, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China

Richard Evans

3 College of Engineering, Design and Physical Sciences, Brunel University London, Uxbridge, United Kingdom

Associated Data

The original contributions presented in the study are included in the article/ Supplementary Material , further inquiries can be directed to the corresponding author/s.

Background: Cyberbullying is well-recognized as a severe public health issue which affects both adolescents and children. Most extant studies have focused on national and regional effects of cyberbullying, with few examining the global perspective of cyberbullying. This systematic review comprehensively examines the global situation, risk factors, and preventive measures taken worldwide to fight cyberbullying among adolescents and children.

Methods: A systematic review of available literature was completed following PRISMA guidelines using the search themes “cyberbullying” and “adolescent or children”; the time frame was from January 1st, 2015 to December 31st, 2019. Eight academic databases pertaining to public health, and communication and psychology were consulted, namely: Web of Science, Science Direct, PubMed, Google Scholar, ProQuest, Communication & Mass Media Complete, CINAHL, and PsycArticles. Additional records identified through other sources included the references of reviews and two websites, Cyberbullying Research Center and United Nations Children's Fund. A total of 63 studies out of 2070 were included in our final review focusing on cyberbullying prevalence and risk factors.

Results: The prevalence rates of cyberbullying preparation ranged from 6.0 to 46.3%, while the rates of cyberbullying victimization ranged from 13.99 to 57.5%, based on 63 references. Verbal violence was the most common type of cyberbullying. Fourteen risk factors and three protective factors were revealed in this study. At the personal level, variables associated with cyberbullying including age, gender, online behavior, race, health condition, past experience of victimization, and impulsiveness were reviewed as risk factors. Likewise, at the situational level, parent-child relationship, interpersonal relationships, and geographical location were also reviewed in relation to cyberbullying. As for protective factors, empathy and emotional intelligence, parent-child relationship, and school climate were frequently mentioned.

Conclusion: The prevalence rate of cyberbullying has increased significantly in the observed 5-year period, and it is imperative that researchers from low and middle income countries focus sufficient attention on cyberbullying of children and adolescents. Despite a lack of scientific intervention research on cyberbullying, the review also identified several promising strategies for its prevention from the perspectives of youths, parents and schools. More research on cyberbullying is needed, especially on the issue of cross-national cyberbullying. International cooperation, multi-pronged and systematic approaches are highly encouraged to deal with cyberbullying.

Introduction

Childhood and adolescence are not only periods of growth, but also of emerging risk taking. Young people during these periods are particularly vulnerable and cannot fully understand the connection between behaviors and consequences ( 1 ). With peer pressures, the heat of passion, children and adolescents usually perform worse than adults when people are required to maintain self-discipline to achieve good results in unfamiliar situations. Impulsiveness, sensation seeking, thrill seeking, and other individual differences cause adolescents to risk rejecting standardized risk interventions ( 2 ).

About one-third of Internet users in the world are children and adolescents under the age of 18 ( 3 ). Digital technology provide a new form of interpersonal communication ( 4 ). However, surveys and news reports also show another picture in the Internet Age. The dark side of young people's internet usage is that they may bully or suffer from others' bullying in cyberspace. This behavior is also acknowledged as cyberbullying ( 5 ). Based on Olweus's definition, cyberbullying is usually regarded as bullying implemented through electronic media ( 6 , 7 ). Specifically, cyberbullying among children and adolescents can be summarized as the intentional and repeated harm from one or more peers that occurs in cyberspace caused by the use of computers, smartphones and other devices ( 4 , 8 – 12 ). In recent years, new forms of cyberbullying behaviors have emerged, such as cyberstalking and online dating abuse ( 13 – 15 ).

Although cyberbullying is still a relatively new field of research, cyberbullying among adolescents is considered to be a serious public health issue that is closely related to adolescents' behavior, mental health and development ( 16 , 17 ). The increasing rate of Internet adoption worldwide and the popularity of social media platforms among the young people have worsened this situation with most children and adolescents experiencing cyberbullying or online victimization during their lives. The confines of space and time are alleviated for bullies in virtual environments, creating new venues for cyberbullying with no geographical boundaries ( 6 ). Cyberbullying exerts negative effects on many aspects of young people's lives, including personal privacy invasion and psychological disorders. The influence of cyberbullying may be worse than traditional bullying as perpetrators can act anonymously and connect easily with children and adolescents at any time ( 18 ). In comparison with traditional victims, those bullied online show greater levels of depression, anxiety and loneliness ( 19 ). Self-esteem problems and school absenteeism have also proven to be related to cyberbullying ( 20 ).

Due to changes in use and behavioral patterns among the youth on social media, the manifestations and risk factors of cyberbullying have faced significant transformation. Further, as the boundaries of cyberbullying are not limited by geography, cyberbullying may not be a problem contained within a single country. In this sense, cyberbullying is a global problem and tackling it requires greater international collaboration. The adverse effects caused by cyberbullying, including reduced safety, lower educational attainment, poorer mental health and greater unhappiness, led UNICEF to state that “no child is absolutely safe in the digital world” ( 3 ).

Extant research has examined the prevalence and risk factors of cyberbullying to unravel the complexity of cyberbullying across different countries and their corresponding causes. However, due to variations in cyberbullying measurement and methodologies, no consistent conclusions have been drawn ( 21 ). Studies into inconsistencies in prevalence rates of cyberbullying, measured in the same country during the same time period, occur frequently. Selkie et al. systematically reviewed cyberbullying among American middle and high school students aged 10–19 years old in 2015, and revealed that the prevalence of cyberbullying victimization ranged from 3 to 72%, while perpetration ranged from 1 to 41% ( 22 ). Risk and protective factors have also been broadly studied, but confirmation is still needed of those factors which have more significant effects on cyberbullying among young people. Clarification of these issues would be useful to allow further research to recognize cyberbullying more accurately.

This review aims to extend prior contributions and provide a comprehensive review of cyberbullying of children and adolescents from a global perspective, with the focus being on prevalence, associated risk factors and protective factors across countries. It is necessary to provide a global panorama based on research syntheses to fill the gaps in knowledge on this topic.

Search Strategies

This study strictly employed Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. We consulted eight academic databases pertaining to public health, and communication and psychology, namely: Web of Science, Science Direct, PubMed, Google Scholar, ProQuest, Communication & Mass Media Complete, CINAHL, and PsycArticles. Additional records identified through other sources included the references of reviews and two websites, Cyberbullying Research Center and United Nations Children's Fund. With regard to the duration of our review, since most studies on cyberbullying arose around 2015 ( 9 , 21 ), this study highlights the complementary aspects of the available information about cyberbullying during the recent 5 year period from January 1st, 2015 to December 31st, 2019.

One researcher extracted keywords and two researchers proposed modifications. We used two sets of subject terms to review articles, “cyberbullying” and “child OR adolescent.” Some keywords that refer to cyberbullying behaviors and young people are also included, such as threat, harass, intimidate, abuse, insult, humiliate, condemn, isolate, embarrass, forgery, slander, flame, stalk, manhunt, as well as teen, youth, young people and student. The search formula is (cyberbullying OR cyber-bullying OR cyber-aggression OR ((cyber OR online OR electronic OR Internet) AND (bully * OR aggres * OR violence OR perpetrat * OR victim * OR threat * OR harass * OR intimidat * OR * OR insult * OR humiliate * OR condemn * OR isolate * OR embarrass * OR forgery OR slander * OR flame OR stalk * OR manhunt))) AND (adolescen * OR child OR children OR teen? OR teenager? OR youth? OR “young people” OR “elementary school student * ” OR “middle school student * ” OR “high school student * ”). The main search approach is title search. Search strategies varied according to the database consulted, and we did not limit the type of literature for inclusion. Journals, conference papers and dissertations are all available.

Specifically, the inclusion criteria for our study were as follows: (a). reported or evaluated the prevalence and possible risk factors associated with cyberbullying, (b). respondents were students under the age of 18 or in primary, junior or senior high schools, and (c). studies were written in English. Exclusion criteria were: (a). respondents came from specific groups, such as clinical samples, children with disabilities, sexual minorities, specific ethnic groups, specific faith groups or samples with cross-national background, (b). review studies, qualitative studies, conceptual studies, book reviews, news reports or abstracts of meetings, and (c). studies focused solely on preventive measures that were usually meta-analytic and qualitative in nature. Figure 1 presents the details of the employed screening process, showing that a total of 63 studies out of 2070 were included in our final review.

PRISMA flow chart diagram showing the process of study selection for inclusion in the systematic review on children and adolescents cyberbullying.

Meta-analysis was not conducted as the limited research published within the 5 years revealed little research which reported odds ratio. On the other hand, due to the inconsistency of concepts, measuring instruments and recall periods, considerable variation could be found in research quality ( 23 ). Meta-analysis is not a preferred method.

Coding Scheme

For coding, we created a comprehensive code scheme to include the characteristics. For cyberbullying, we coded five types proposed by Willard ( 24 – 26 ), which included verbal violence, group violence, visual violence, impersonating and account forgery, and other behaviors. Among them, verbal violence is considered one of the most common types of cyberbullying and refers to the behavior of offensive responses, insults, mocking, threats, slander, and harassment. Group violence is associated with preventing others from joining certain groups or isolating others, forcing others to leave the group. Visual violence relates to the release and sharing of embarrassing photos and information without the owners' consent. Impersonating and account forgery refers to identity theft, stealing passwords, violating accounts and the creation of fake accounts to fraudulently present the behavior of others. Other behaviors include disclosure of privacy, sexual harassment, and cyberstalking. To comprehensively examine cyberbullying, we coded cyberbullying behaviors from both the perspectives of cyberbullying perpetrators and victims, if mentioned in the studies.

In relation to risk factors, we drew insights from the general aggression model, which contributes to the understanding of personal and situational factors in the cyberbullying of children and adolescents. We chose the general aggression model because (a) it contains more situational factors than other models (e.g., social ecological models) - such as school climate ( 9 ), and (b) we believe that the general aggression model is more suitable for helping researchers conduct a systematic review of cyberbullying risk and protective factors. This model provides a comprehensive framework that integrates domain specific theories of aggression, and has been widely applied in cyberbullying research ( 27 ). For instance, Kowalski and colleagues proposed a cyberbullying encounter through the general aggression model to understand the formation and development process of youth cyberbullying related to both victimization and perpetration ( 9 ). Victims and perpetrators enter the cyberbullying encounter with various individual characteristics, experiences, attitudes, desires, personalities, and motives that intersect to determine the course of the interaction. Correspondingly, the antecedents pertaining to cyberbullying are divided into two broad categories, personal factors and situational factors. Personal factors refer to individual characteristics, such as gender, age, motivation, personality, psychological states, socioeconomic status and technology use, values and perceptions, and other maladaptive behaviors. Situational factors focus on the provocation/support, parental involvement, school climate, and perceived anonymity. Consequently, our coders related to risk factors consisting of personal factors and situational factors from the perspectives of both cyberbullying perpetrators and victims.

We extracted information relating to individual papers and sample characteristics, including authors, year of publication, country, article type, sampling procedures, sample characteristics, measures of cyberbullying, and prevalence and risk factors from both cyberbullying perpetration and victimization perspectives. The key words extraction and coding work were performed twice by two trained research assistants in health informatics. The consistency test results are as follows: the Kappa value with “personal factors” was 0.932, and the Kappa value with “situational factors” was 0.807. The result shows that the coding consistency was high enough and acceptable. Disagreements were resolved through discussion with other authors.

Quality Assessment of Studies

The quality assessment of the studies is based on the recommended tool for assessing risk of bias, Cochrane Collaboration. This quality assessment tool focused on seven items: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other sources of bias ( 28 ). We assessed each item as “low risk,” “high risk,” and “unclear” for included studies. A study is considered of “high quality” when it meets three or more “low risk” requirements. When one or more main flaw of a study may affect the research results, the study is considered as “low quality.” When a lack of information leads to a difficult judgement, the quality is considered to be “unclear.” Please refer to Appendix 1 for more details.

This comprehensive systematic review comprised a total of 63 studies. Appendices 2 , 3 show the descriptive information of the studies included. Among them, 58 (92%) studies measured two or more cyberbullying behavior types. The sample sizes of the youths range from several hundred to tens of thousands, with one thousand to five thousand being the most common. As for study distribution, the United States of America, Spain and China were most frequently mentioned. Table 1 presents the detail.

Descriptive information of studies included (2015–2019).

Prevalence of Global Cyberbullying

Prevalence across countries.

Among the 63 studies included, 22 studies reported on cyberbullying prevalence and 20 studies reported on prevalence from victimization and perpetration perspectives, respectively. Among the 20 studies, 11 national studies indicated that the prevalence of cyberbullying victimization and cyberbullying perpetration ranged from 14.6 to 52.2% and 6.3 to 32%, respectively. These studies were conducted in the United States of America ( N = 4) ( 29 – 32 ), South Korea ( N = 3) ( 33 – 35 ), Singapore ( N = 1) ( 36 ), Malaysia ( N = 1) ( 37 ), Israel ( N = 1) ( 38 ), and Canada ( N = 1) ( 39 ). Only one of these 11 national studies is from an upper middle income country, and the rest are from highincome countries identified by the World Bank ( 40 ). By combining regional and community-level studies, the prevalence of cyberbullying victimization and cyberbullying perpetration ranged from 13.99 to 57.5% and 6.0 to 46.3%, respectively. Spain reported the highest prevalence of cyberbullying victimization (57.5%) ( 41 ), followed by Malaysia (52.2%) ( 37 ), Israel (45%) ( 42 ), and China (44.5%) ( 43 ). The lowest reported victim rates were observed in Canada (13.99%) and South Korea (14.6%) ( 34 , 39 ). The reported prevalence of cyberbullying victimization in the United States of America ranged from 15.5 to 31.4% ( 29 , 44 ), while in Israel, rates ranged from 30 to 45% ( 26 , 42 ). In China, rates ranged from 6 to 46.3% with the country showing the highest prevalence of cyberbullying perpetration (46.30%) ( 15 , 43 , 45 , 46 ). Canadian and South Korean studies reported the lowest prevalence of cyberbullying perpetration at 7.99 and 6.3%, respectively ( 34 , 39 ).

A total of 10 studies were assessed as high quality studies. Among them, six studies came from high income countries, including Canada, Germany, Italy, Portugal, and South Korea ( 13 , 34 , 39 , 46 – 48 ). Three studies were from upper middle income countries, including Malaysia and China ( 37 , 43 ) and one from a lower middle income country, Nigeria ( 49 ). Figures 2 , ​ ,3 3 describe the prevalence of cyberbullying victimization and perpetration respectively among high quality studies.

The prevalence of cyberbullying victimization of high quality studies.

The prevalence of cyberbullying perpetration of high quality studies.

Prevalence of Various Cyberbullying Behaviors

For the prevalence of cyberbullying victimization and perpetration, the data were reported in 18 and 14 studies, respectively. Figure 4 shows the distribution characteristics of the estimated value of prevalence of different cyberbullying behaviors with box plots. The longer the box, the greater the degree of variation of the numerical data and vice versa. The rate of victimization and crime of verbal violence, as well as the rate of victimization of other behaviors, such as cyberstalking and digital dating abuse, has a large degree of variation. Among the four specified types of cyberbullying behaviors, verbal violence was regarded as the most commonly reported behaviors in both perpetration and victimization rates, with a wide range of prevalence, ranging from 5 to 18%. Fewer studies reported the prevalence data for visual violence and group violence. Studies also showed that the prevalence of impersonation and account forgery were within a comparatively small scale. Specific results were as follows.

Cyberbullying prevalence across types (2015–2019).

Verbal Violence

A total of 13 studies reported verbal violence prevalence data ( 15 , 26 , 34 , 37 – 39 , 42 , 43 , 47 , 48 , 50 , 51 ). Ten studies reported the prevalence of verbal violence victimization ranging from 2.8 to 47.5%, while seven studies claimed perpetration prevalence ranging from 1.5 to 31.8%. Malaysia reported the highest prevalence of verbal violence victimization (47.5%) ( 37 ), followed by China (32%) ( 43 ). China reported that the prevalence of verbal violence victimization ranged from 5.1 to 32% ( 15 , 43 ). Israel reported that the prevalence of verbal violence victimization ranged from 3.4 to 18% ( 26 , 38 , 42 ). For perpetration rate, Malaysia reported the highest level at 31.8% ( 37 ), while a study for Spain reported the lowest, ranging from 3.2 to 6.4% ( 51 ).

Group Violence

The prevalence of group violence victimization was explored within 4 studies and ranged from 5 to 17.8% ( 26 , 34 , 42 , 43 ), while perpetration prevalence was reported in three studies, ranging from 10.1 to 19.07% ( 34 , 43 , 47 ). An Israeli study suggested that 9.8% of respondents had been excluded from the Internet, while 8.9% had been refused entry to a group or team ( 26 ). A study in South Korea argued that the perpetration prevalence of group violence was 10.1% ( 34 ), while a study in Italy reported that the rate of online group violence against others was 19.07% ( 47 ).

Visual Violence

The prevalence of visual violence victimization was explored within three studies and ranged from 2.6 to 12.1% ( 26 , 34 , 43 ), while the perpetration prevalence reported in four studies ranged from 1.7 to 6% ( 34 , 43 , 47 , 48 ). For victimization prevalence, a South Korean study found that 12.1% of respondents reported that their personal information was leaked online ( 34 ). An Israel study reported that the prevalence of outing the picture was 2.6% ( 26 ). For perpetration prevalence, a South Korean study found that 1.7% of respondents had reported that they had disclosed someone's personal information online ( 34 ). A German study reported that 6% of respondents had written a message (e.g., an email) to somebody using a fake identity ( 48 ).

Impersonating and Account Forgery

Four studies reported on the victimization prevalence of impersonating and account forgery, ranging from 1.1 to 10% ( 15 , 42 , 43 ), while five studies reported on perpetration prevalence, with the range being from 1.3 to 9.31% ( 15 , 43 , 47 , 48 , 51 ). In a Spanish study, 10% of respondents reported that their accounts had been infringed by others or that they could not access their account due to stolen passwords. In contrast, 4.5% of respondents reported that they had infringed other people's accounts or stolen passwords, with 2.5% stating that they had forged other people's accounts ( 51 ). An Israeli study reported that the prevalence of being impersonated was 7% ( 42 ), while in China, a study reported this to be 8.6% ( 43 ). Another study from China found that 1.1% of respondents had been impersonated to send dating-for-money messages ( 15 ).

Other Behaviors

The prevalence of disclosure of privacy, sexual harassment, and cyberstalking were also explored by scholars. Six studies reported the victimization prevalence of other cyberbullying behaviors ( 13 , 15 , 34 , 37 , 42 , 43 ), and four studies reported on perpetration prevalence ( 34 , 37 , 43 , 48 ). A study in China found that 1.2% of respondents reported that their privacy had been compromised without permission due to disputes ( 15 ). A study from China reported the prevalence of cyberstalking victimization was 11.9% ( 43 ), while a Portuguese study reported that this was 62% ( 13 ). In terms of perpetration prevalence, a Malaysian study reported 2.7% for sexual harassment ( 37 ).

Risk and Protective Factors of Cyberbullying

In terms of the risk factors associated with cyberbullying among children and adolescents, this comprehensive review highlighted both personal and situational factors. Personal factors referred to age, gender, online behavior, race, health conditions, past experiences of victimization, and impulsiveness, while situational factors consisted of parent-child relationship, interpersonal relationships, and geographical location. In addition, protective factors against cyberbullying included: empathy and emotional intelligence, parent-child relationship, and school climate. Table 2 shows the risk and protective factors for child and adolescent cyberbullying.

Risk and protective factors of cyberbullying among children and adolescents.

In terms of the risk factors associated with cyberbullying victimization at the personal level, many studies evidenced that females were more likely to be cyberbullied than males ( 13 , 26 , 29 , 38 , 43 , 52 , 54 , 55 , 58 ). Meanwhile, adolescents with mental health problems ( 61 ), such as depression ( 33 , 62 ), borderline personality disorder ( 63 ), eating disorders ( 41 ), sleep deprivation ( 56 ), and suicidal thoughts and suicide plans ( 64 ), were more likely to be associated with cyberbullying victimization. As for Internet usage, researchers agreed that youth victims were probably those that spent more time online than their counterparts ( 32 , 36 , 43 , 45 , 48 , 49 , 60 ). For situational risk factors, some studies have proven the relationship between cyberbullying victims and parental abuse, parental neglect, family dysfunction, inadequate monitoring, and parents' inconsistency in mediation, as well as communication issues ( 33 , 64 , 68 , 73 ). In terms of geographical location, some studies have reported that youths residing in city locations are more likely to be victims of cyberbullying than their peers from suburban areas ( 61 ).

Regarding the risk factors of cyberbullying perpetration at the personal level, it is generally believed that older teenagers, especially those aged over 15 years, are at greater risk of becoming cyberbullying perpetrators ( 55 , 67 ). When considering prior cyberbullying experiences, evidence showed that individuals who had experienced cyberbullying or face-to-face bullying tended to be aggressors in cyberbullying ( 35 , 42 , 49 , 51 , 55 ); in addition, the relationship between impulsiveness and cyberbullying perpetration was also explored by several pioneering scholars ( 55 , 72 , 80 ). The situational factors highlight the role of parents and teachers in cyberbullying experiences. For example, over-control and authoritarian parenting styles, as well as inharmonious teacher-student relationships ( 61 ) are perceived to lead to cyberbullying behaviors ( 74 , 75 ). In terms of differences in geographical locations, students residing in cities have a higher rate of online harassment than students living in more rural locations ( 49 ).

In terms of the protective factors in child and adolescent cyberbullying, scholars have focused on youths who have limited experiences of cyberbullying. At the personal level, high emotional intelligence, an ability for emotional self-control and empathy, such as cognitive empathy ability ( 44 , 55 ), were associated with lower rates of cyberbullying ( 57 ). At the situational level, a parent's role is seen as critical. For example, intimate parent-child relationships ( 46 ) and open active communication ( 19 ) were demonstrated to be related to lower experiences of cyberbullying and perpetration. Some scholars argued that parental supervision and monitoring of children's online activities can reduce their tendency to participate in some negative activities associated with cyberbullying ( 31 , 46 , 73 ). They further claimed that an authoritative parental style protects youths against cyberbullying ( 43 ). Conversely, another string of studies evidenced that parents' supervision of Internet usage was meaningless ( 45 ). In addition to conflicting roles of parental supervision, researchers have also looked into the role of schools, and posited that positive school climates contribute to less cyberbullying experiences ( 61 , 79 ).

Some risk factors may be protective factors under another condition. Some studies suggest that parental aggressive communication is related to severe cyberbullying victims, while open communication is a potential protective factor ( 19 ). Parental neglect, parental abuse, parental inconsistency in supervision of adolescents' online behavior, and family dysfunction are related to the direct or indirect harm of cyberbullying ( 33 , 68 ). Parental participation, a good parental-children relationship, communication and dialogue can enhance children's school adaptability and prevent cyberbullying behaviors ( 31 , 74 ). When parental monitoring reaches a balance between control and openness, it could become a protective factor against cyberbullying, and it could be a risk factor, if parental monitoring is too low or over-controlled ( 47 ).

Despite frequent discussion about the risk factors associated with cyberbullying among children and adolescents, some are still deemed controversial factors, such as age, race, gender, and the frequency of suffering on the internet. For cyberbullying victims, some studies claim that older teenagers are more vulnerable to cyberbullying ( 15 , 38 , 52 , 53 ), while other studies found conflicting results ( 26 , 33 ). As for student race, Alhajji et al. argued that non-white students were less likely to report cyberbullying ( 29 ), while Morin et al. observed no significant correlation between race and cyberbullying ( 52 ). For cyberbullying perpetration, Alvarez-Garcia found that gender differences may have indirect effects on cyberbullying perpetration ( 55 ), while others disagreed ( 42 , 61 , 68 – 70 ). Specifically, some studies revealed that males were more likely to become cyberbullying perpetrators ( 34 , 39 , 56 ), while Khurana et al. presented an opposite point of view, proposing that females were more likely to attack others ( 71 ). In terms of time spent on the Internet, some claimed that students who frequently surf the Internet had a higher chance of becoming perpetrators ( 49 ), while others stated that there was no clear and direct association between Internet usage and cyberbullying perpetration ( 55 ).

In addition to personal and situational factors, scholars have also explored other specific factors pertaining to cyberbullying risk and protection. For instance, mindfulness and depression were found to be significantly related to cyber perpetration ( 76 ), while eating disorder psychopathology in adolescents was associated with cyber victimization ( 41 ). For males who were familiar with their victims, such as family members, friends and acquaintances, they were more likely to be cyberstalking perpetrators than females or strangers, while pursuing desired closer relationships ( 13 ). In the school context, a lower social likability in class was identified as an indirect factor for cyberbullying ( 48 ).

This comprehensive review has established that the prevalence of global childhood and adolescent victimization from cyberbullying ranges from 13.99 to 57.5%, and that the perpetration prevalence ranges from 6.0 to 46.3%. Across the studies included in our research, verbal violence is observed as one of the most common acts of cyberbullying, including verbal offensive responses, insults, mocking, threats, slander, and harassment. The victimization prevalence of verbal violence is reported to be between 5 and 47.5%, and the perpetration prevalence is between 3.2 and 26.1%. Personal factors, such as gender, frequent use of social media platforms, depression, borderline personality disorder, eating disorders, sleep deprivation, and suicidal tendencies, were generally considered to be related to becoming a cyberbullying victim. Personal factors, such as high school students, past experiences, impulse, improperly controlled family education, poor teacher-student relationships, and the urban environment, were considered risk factors for cyberbullying perpetration. Situational factors, including parental abuse and neglect, improper monitoring, communication barriers between parents and children, as well as the urban environment, were also seen to potentially contribute to higher risks of both cyberbullying victimization and perpetration.

Increasing Prevalence of Global Cyberbullying With Changing Social Media Landscape and Measurement Alterations

This comprehensive review suggests that global cyberbullying rates, in terms of victimization and perpetration, were on the rise during the 5 year period, from 2015 to 2019. For example, in an earlier study conducted by Modecki et al. the average cyberbullying involvement rate was 15% ( 81 ). Similar observations were made by Hamm et al. who found that the median rates of youth having experienced bullying or who had bullied others online, was 23 and 15.2%, respectively ( 82 ). However, our systematic review summarized global children and adolescents cyberbullying in the last 5 years and revealed an average cyberbullying perpetration rate of 25.03%, ranging from 6.0 to 46.3%, while the average victimization was 33.08%, ranging from 13.99 to 57.5%. The underlying reason for increases may be attributed to the rapid changing landscape of social media and, in recent years, the drastic increase in Internet penetration rates. With the rise in Internet access, youths have greater opportunities to participate in online activities, provided by emerging social media platforms.

Although our review aims to provide a broader picture of cyberbullying, it is well-noted in extant research that difficulties exist in accurately estimating variations in prevalence in different countries ( 23 , 83 ). Many reasons exist to explain this. The first largely relates poor or unclear definition of the term cyberbullying; this hinders the determination of cyberbullying victimization and perpetration ( 84 ). Although traditional bullying behavior is well-defined, the definition cannot directly be applied to the virtual environment due to the complexity in changing online interactions. Without consensus on definitions, measurement and cyberbullying types may vary noticeably ( 83 , 85 ). Secondly, the estimation of prevalence of cyberbullying is heavily affected by research methods, such as recall period (lifetime, last year, last 6 months, last month, or last week etc.), demographic characteristics of the survey sample (age, gender, race, etc.), perspectives of cyberbullying experiences (victims, perpetrators, or both victim and perpetrator), and instruments (scales, study-specific questions) ( 23 , 84 , 86 ). The variety in research tools and instruments used to assess the prevalence of cyberbullying can cause confusion on this issue ( 84 ). Thirdly, variations in economic development, cultural backgrounds, human values, internet penetration rates, and frequency of using social media may lead to different conclusions across countries ( 87 ).

Acknowledging the Conflicting Role of the Identified Risk Factors With More Research Needed to Establish the Causality

Although this review has identified many personal and situational factors associated with cyberbullying, the majority of studies adopted a cross-sectional design and failed to reveal the causality ( 21 ). Nevertheless, knowledge on these correlational relationships provide valuable insights for understanding and preventing cyberbullying incidents. In terms of gender differences, females are believed to be at a higher risk of cyberbullying victimization compared to males. Two reasons may help to explain this. First, the preferred violence behaviors between two genders. females prefer indirect harassment, such as the spreading of rumors, while males tend toward direct bullying (e.g., assault) ( 29 ) and second, the cultural factors. From the traditional gender perspective, females tended to perceive a greater risk of communicating with others on the Internet, while males were more reluctant to express fear, vulnerability and insecurity when asked about their cyberbullying experiences ( 46 ). Females were more intolerant when experiencing cyberstalking and were more likely to report victimization experiences than males ( 13 ). Meanwhile, many researchers suggested that females are frequent users of emerging digital communication platforms, which increases their risk of unpleasant interpersonal contact and violence. From the perspective of cultural norms and masculinity, the reporting of cyberbullying is also widely acknowledged ( 37 ). For example, in addition, engaging in online activities is also regarded as a critical predictor for cyberbullying victimization. Enabled by the Internet, youths can easily find potential victims and start harassment at any time ( 49 ). Participating in online activities directly increases the chance of experiencing cyberbullying victimization and the possibility of becoming a victim ( 36 , 45 ). As for age, earlier involvement on social media and instant messaging tools may increase the chances of experiencing cyberbullying. For example, in Spain, these tools cannot be used without parental permission before the age of 14 ( 55 ). Besides, senior students were more likely to be more impulsive and less sympathetic. They may portray more aggressive and anti-social behaviors ( 55 , 72 ); hence senior students and students with higher impulsivity were usually more likely to become cyberbullying perpetrators.

Past experiences of victimization and family-related factors are another risk for cyberbullying crime. As for past experiences, one possible explanation is that young people who had experienced online or traditional school bullying may commit cyberbullying using e-mails, instant messages, and text messages for revenge, self-protection, or improving their social status ( 35 , 42 , 49 , 55 ). In becoming a cyberbullying perpetrator, the student may feel more powerful and superior, externalizing angry feelings and relieving the feelings of helplessness and sadness produced by past victimization experiences ( 51 ). As for family related factors, parenting styles are proven to be highly correlated to cyberbullying. In authoritative families, parents focus on rational behavioral control with clear rules and a high component of supervision and parental warmth, which have beneficial effects on children's lifestyles ( 43 ). Conversely, in indulgent families, children's behaviors are not heavily restricted and parents guide and encourage their children to adapt to society. The characteristics of this indulgent style, including parental support, positive communication, low imposition, and emotional expressiveness, possibly contribute to more parent-child trust and less misunderstanding ( 75 ). The protective role of warmth/affection and appropriate supervision, which are common features of authoritative or indulgent parenting styles, mitigate youth engagement in cyberbullying. On the contrary, authoritarian and neglectful styles, whether with excessive or insufficient control, are both proven to be risk factors for being a target of cyberbullying ( 33 , 76 ). In terms of geographical location, although several studies found that children residing in urban areas were more likely to be cyberbullying victims than those living in rural or suburban areas, we cannot draw a quick conclusion here, since whether this difference attributes to macro-level differences, such as community safety or socioeconomic status, or micro-level differences, such as teacher intervention in the classroom, courses provided, teacher-student ratio, is unclear across studies ( 61 ). An alternative explanation for this is the higher internet usage rate in urban areas ( 49 ).

Regarding health conditions, especially mental health, some scholars believe that young people with health problems are more likely to be identified as victims than people without health problems. They perceive health condition as a risk factor for cyberbullying ( 61 , 63 ). On the other hand, another group of scholars believe that cyberbullying has an important impact on the mental health of adolescents which can cause psychological distress consequences, such as post-traumatic stress mental disorder, depression, suicidal ideation, and drug abuse ( 70 , 87 ). It is highly possible that mental health could be risk factors, consequences of cyberbullying or both. Mental health cannot be used as standards, requirements, or decisive responses in cyberbullying research ( 13 ).

The Joint Effort Between Youth, Parents, Schools, and Communities to Form a Cyberbullying-Free Environment

This comprehensive review suggests that protecting children and adolescents from cyberbullying requires joint efforts between individuals, parents, schools, and communities, to form a cyberbullying-free environment. For individuals, young people are expected to improve their digital technology capabilities, especially in the use of social media platforms and instant messaging tools ( 55 ). To reduce the number of cyberbullying perpetrators, it is necessary to cultivate emotional self-regulation ability through appropriate emotional management training. Moreover, teachers, counselors, and parents are required to be armed with sufficient knowledge of emotional management and to develop emotional management capabilities and skills. In this way, they can be alert to the aggressive or angry emotions expressed by young people, and help them mediate any negative emotions ( 45 ), and avoid further anti-social behaviors ( 57 ).

For parents, styles of parenting involving a high level of parental involvement, care and support, are desirable in reducing the possibility of children's engagement in cyberbullying ( 74 , 75 ). If difficulties are encountered, open communication can contribute to enhancing the sense of security ( 73 ). In this vein, parents should be aware of the importance of caring, communicating and supervising their children, and participate actively in their children's lives ( 71 ). In order to keep a balance between control and openness ( 47 ), parents can engage in unbiased open communication with their children, and reach an agreement on the usage of computers and smart phones ( 34 , 35 , 55 ). Similarly, it is of vital importance to establish a positive communication channel with children ( 19 ).

For schools, a higher priority is needed to create a safe and positive campus environment, providing students with learning opportunities and ensuring that every student is treated equally. With a youth-friendly environment, students are able to focus more on their academic performance and develop a strong sense of belonging to the school ( 79 ). For countries recognizing collectivist cultural values, such as China and India, emphasizing peer attachment and a sense of collectivism can reduce the risk of cyberbullying perpetration and victimization ( 78 ). Besides, schools can cooperate with mental health agencies and neighboring communities to develop preventive programs, such as extracurricular activities and training ( 44 , 53 , 62 ). Specifically, school-based preventive measures against cyberbullying are expected to be sensitive to the characteristics of young people at different ages, and the intersection of race and school diversity ( 29 , 76 ). It is recommended that school policies that aim to embrace diversity and embody mutual respect among students are created ( 26 ). Considering the high prevalence of cyberbullying and a series of serious consequences, it is suggested that intervention against cyberbullying starts from an early stage, at about 10 years old ( 54 ). Schools can organize seminars to strengthen communication between teachers and students so that they can better understand the needs of students ( 61 ). In addition, schools should encourage cyberbullying victims to seek help and provide students with opportunities to report cyberbullying behaviors, such as creating online anonymous calls.

Conclusions and Limitations

The comprehensive study has reviewed related research on children and adolescents cyberbullying across different countries and regions, providing a positive understanding of the current situation of cyberbullying. The number of studies on cyberbullying has surged in the last 5 years, especially those related to risk factors and protective factors of cyberbullying. However, research on effective prevention is insufficient and evaluation of policy tools for cyberbullying intervention is a nascent research field. Our comprehensive review concludes with possible strategies for cyberbullying prevention, including personal emotion management, digital ability training, policy applicability, and interpersonal skills. We highlight the important role of parental control in cyberbullying prevention. As for the role of parental control, it depends on whether children believe their parents are capable of adequately supporting them, rather than simply interfering in their lives, restricting their online behavior, and controlling or removing their devices ( 50 ). In general, cyberbullying is on the rise, with the effectiveness of interventions to meet this problem still requiring further development and exploration ( 83 ).

Considering the overlaps between cyberbullying and traditional offline bullying, future research can explore the unique risk and protective factors that are distinguishable from traditional bullying ( 86 ). To further reveal the variations, researchers can compare the outcomes of interventions conducted in cyberbullying and traditional bullying preventions simultaneously, and the same interventions only targeting cyberbullying ( 88 ). In addition, cyberbullying also reflects a series of other social issues, such as personal privacy and security, public opinion monitoring, multinational perpetration and group crimes. To address this problem, efforts from multiple disciplines and novel analytical methods in the digital era are required. As the Internet provides enormous opportunities to connect young people from all over the world, cyberbullying perpetrators may come from transnational networks. Hence, cyberbullying of children and adolescents, involving multiple countries, is worth further attention.

Our study has several limitations. First, national representative studies are scarce, while few studies from middle and low income countries were included in our research due to language restrictions. Many of the studies included were conducted in schools, communities, provinces, and cities in high income countries. Meanwhile, our review only focused on victimization and perpetration. Future studies should consider more perspectives, such as bystanders and those with the dual identity of victim/perpetrator, to comprehensively analyze the risk and protective factors of cyberbullying.

Data Availability Statement

Author contributions.

SH, CZ, RE, and WZ conceived the study and developed the design. WZ analyzed the result and supervised the study. CZ and SH wrote the first draft. All authors contributed to the article and approved the submitted version.

Conflict of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Supplementary Material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpubh.2021.634909/full#supplementary-material

• " class="ru"> " class="language-link">Russian

PhD Thesis: Of all Types of Bullying, Cyberbullying is Most Devastating to Well-Being of Adolescents

On 26 June, Lauraliisa Mark from the Tallinn University School of Governance, Law and Society will defend her doctoral thesis, which studied the suicidal thoughts and depressive feelings among Estonian youth.

According to the study, in the past 12 months, 13% of students in grades 7-9 had had suicidal thoughts, and 26% had felt depression.

According to the World Health Organisation, suicide is one of the main causes of death among 15-29-year-olds. At the same time, among 15-19-year-olds in Estonia, the rate of suicide is one of the highest in the world. “In my thesis, I studied the spread of suicidal thoughts and depressive feelings among the pupils in grades 5-9, and analysed the relation between suicidal thoughts, depressive feelings, and risk behaviour,” Mark explained and added that depression is one of the most common mental disorders to lead to suicide.

“Even though risk behaviour is often seen as a normal part of being young, all risk behaviours analysed in this thesis (early sexual activity, regular smoking, alcohol use, getting into physical fights, bullying) were potential sources of suicidal ideas and/or depressive feelings,” Mark stated. For example, being sexually active at 13 years old or earlier was four times (among boys) to eight times (among girls) more likely to lead to suicidal thoughts. “The earlier a person became sexually active, the more often they would also consume alcohol, smoke, fight, bully, and have worse estimation toward their health,” she added.

“In the case of multiple risk factors appearing simultaneously among 15-year-olds, the emergence of suicidal thoughts increased multiple times; in the case of four or five factors, the probability increased by more than 10 times,” she said and added that ease of communication with parents, especially mothers, was also an important factor. “When the young person had difficulties talking to their mother, the probability of suicidal thoughts increased five times. The ease of communication with parents was an important buffer in stopping suicidal thoughts,” Mark added.

Both being bullied and bullying was seen as a result of severe disturbances in mental well-being. Among boys, the biggest disturbances in well-being came from being bullied, girls were equally disturbed while being bullied and/or bullying. Among all groups, cyberbullying turned out to be the most devastating form of bullying.

The results show that risk behaviour should be seen as a sign of danger in the mental health of adolescents, not just a trait of their age. Especially in cases when risk behaviour tends to cumulate. “The risk behaviour of young people can and should be used to recognise and prevent possible mental health problems. In addition to minimising risk behaviour, bullying in schools must also be contained. Research-based methods of school bullying must be applied as soon as possible, continuously, and they should incorporate all parties of bullying (including onlookers),” Mark said.

The PhD thesis “Depressive Feelings and Suicidal Ideation Among Estonian Adolescents and Associations with Selected Risk Behaviours” was defended on 26 June at Tallinn University. The supervisors were Professors Airi Värnik and Merike Sisask from Tallinn University. The opponents were Professor Rory O’Connor from University of Glasgow and Associate Professor Dagmar Kutsar from University of Tartu.

You can access the full text of the thesis via the Tallinn University Academic Library e-vault ETERA .

Related news

Press Release

Support groups for loved ones of people diagnosed with dementia are helpful in getting necessary information and social support for care

Merle Varik, from the School of Governance, Law and Society, defended her doctoral thesis on 22 February 2022. The thesis is entitled "Empowering Informal Caregivers of People with Dementia Through Su...

Doctoral studies

Doctoral dissertation: Participation in adult learning requires multilevel support – important is the role of the individual, the workplace as well as the country-specific institutions

On 28 January 2022 at 17.00, Eve-Liis Roosmaa from the School of Governance, Law and Society will defend her doctoral thesis “Adult education and training: a comparative perspective of participation p...

Doctoral thesis explores feminist threads in Johannes Semper’s oeuvre

On 24 January 2022 at 14.30, Merlin Kirikal from Tallinn University School of Humanities defends her doctoral thesis ““I broke open to life”: representing modern gender and bodies in Johannes Semper’s...

The results of doctoral thesis mapped coping mechanisms for academic burnout among middle school students in Estonia

At 10:00 on 16 December 2021, Kati Vinter, doctoral student of Tallinn University’s School of Educational Sciences, will defend her thesis “Academic burnout among middle-school students in Estonia: An...

Doctoral dissertation: The mental health of children is influenced by the context and the ratings affected by the subjectivity of the assessor

At 16:00 on 15 December 2021, Karin Streimann from Tallinn University’s School of Governance, Law and Society will defend her doctoral thesis “Preventing Mental Health Problems in Elementary Schools: ...

Doctoral thesis: Research into creative thinking is vital for preservation of humanity

At 12 o’clock on 14 December 2021, Elen Lotman of Baltic Film, Media and Arts School will defend her doctoral thesis, "Experiential Heuristics in Fiction Film Cinematography“ („Kogemuslikud heuristika...

• MyAucklandUni
• Student Services Online
• Class search
• Student email
• Change my password
• MyCDES+ (job board)
• Course outlines
• Learning essentials
• Libraries and Learning Services
• Forms, policies and guidelines
• Campus Card
• Enrol in courses
• Postgraduate students
• Summer school
• AskAuckland
• Student Hubs
• Student IT Hub
• Student Health and Counselling
• Harassment, bullying, sexual assault and other violence
• Complaints and incidents
• Career Development and Employability Services (CDES)
• Ratonga Hauātanga Tauira | Student Disability Services (SDS)
• Rainbow support
• Covid-19 information for our community
• Emergency information
• Report concerns, incidents and hazards
• Health and safety topics
• Staff email
• Staff intranet
• ResearchHub
• PeopleSoft HR
• Forms register
• Careers at the University
• Education Office
• Early childhood centres
• University Calendar
• Opportunities
• Update your details
• Make a donation
• Publications
• Photo galleries
• Video and audio
• Career services
• Virtual Book Club
• Library services
• Alumni benefits
• Office contact details
• Alumni and friends on social media
• No events scheduled for today You have no more events scheduled for today
• Next event:
• Show {0} earlier events Show {0} earlier event
• Event_Time Event_Name Event_Description
• My Library Account
• Change Password
• Edit Profile
• My GPA Grade Point Average About your GPA GPA not available Why can't I see my GPA?
• My Progress
• Points Required Completed points My Progress Progress not available All done!
• Student hubs
• Health and counselling
• All support
• Health, safety and well-being

Breadcrumbs List.

• News and opinion

Awakening associations with Gretchen Albrecht

27 May 2024

Ingenio , Alumni , Faculty of Arts

Luke Smythe drew on connections to Gretchen Albrecht that stretched back to his childhood for his extensive book on the artist.

You could argue Dr Luke Smythe started researching his first book when he was nine years old.

It was the mid-1980s and his parents took him and his cousin to a studio on Auckland’s Quay St to look at some paintings by their friend Gretchen Albrecht.

“My parents asked us for our opinions, and I pointed to one and I remember just going ‘I like those colours’. And my parents got it.

The work – two huge, glowing quarter-circle panels, one purple, one green, butting together to form a distinctively Albrechtian hemisphere – became part of Smythe’s childhood. It was called Colloquy (2) (Reflection) .

“I can still see the painting kind of floating on the wall in the dining room.” Three-and-a-bit decades later, the artwork also ended up in Gretchen Albrecht: Between Gesture and Geometry , an extensively illustrated monograph in which Luke traces his friend’s life and work: inspiration in a high-school art class; studies at Elam art school; the critical and commercial success that continues today, in Albrecht’s 81st year.

Now the 2019 book is back in a revised edition that expands on the 1970s section (“Gretchen happened to find a number of paintings in storage”) and brings it up to the minute (“She has made a significant amount of work in the past four years”).

Luke is speaking from Berlin, where he’s researching a book on painter Sigmar Polke, but he’s about to return to Melbourne, where he lectures in art history at Monash University. He has a BA from the University of Auckland, an MA from Columbia and a PhD from Yale: his thesis became the basis of a 2022 book on Gerhard Richter.

The Albrecht book, though, will always be a career pinnacle. Writing it took nine months, but it had been “percolating” for years. In the 2000s, Luke spent years cataloguing her work, “so I probably laid eyes on as many of her works as it’s possible to lay eyes on”.

Albrecht is important as a female artist who achieved great success in a man’s world. What’s striking, though, is the early support she received from male curators and artists, including Hamish Keith, Colin McCahon and Gordon Walters. 

I probably laid eyes on as many of her works as it’s possible to lay eyes on.

Dr Luke Smythe

“That needs to be part of any history of women’s art, right? In a world controlled by men, some women benefit from the assistance of men, who aren’t always out to keep women down. And she succeeded by dint of hard work.”

Albrecht’s art is “very beautiful, and people respond to that”. But Luke also wants to be clear about its importance in the broader history of abstract art.

“By the mid-20th century, there was this movement called Formalism, which maintained a painting was only shapes and colours on the flat surface, and don’t you dare say that they look like something because that would be a Philistine response.”

By the 1970s, says Luke, such ‘pure’ abstraction was catching on in New Zealand. Yet as Albrecht embraced abstraction through the 1970s, her work still held connections to ideas, words and objects in the world – what you might call ‘impure’ abstraction.

Yes, there were landscapes in the horizontal layers of her 1970s ‘stained acrylic’ works; yes, there was a conversation taking place on the Smythe dining-room wall between the two halves of Colloquy (2) .

This was bold. “Gretchen may have been the first person to part company with what had become the orthodoxy of Formalism in New Zealand. She wasn’t being antagonistic; she was just going to do what she felt she wanted to do,” says Luke.

“People couldn’t go on forever pretending that abstract paintings don’t awaken associations, but Gretchen really embraced them. That’s what makes her important.

This article first appeared in the Autumn 2024 edition of Ingenio .