Cyberbullying Classification Methods for Arabic: A Systematic Review

Reem ALBayari, Sherief Abdallah, Said A. Salloum

Abstract: Across the globe, bullying and cyberbullying are present and have significant implications for people and communities. Over the last years, the number of research studies for cyberbullying classification in texts has grown exponentially, but most studies have been examined on English datasets. Most of the new methods in text classification are based on linguistic features of the text since the Arabic language has different rules and structures. This research intended to review articles about cyberbullying classification methods for Arabic texts and analyzed them The review includes the used methods, quality measuring performance, and details of the dataset (dialect, annotation method, source). This will guide future research by giving researchers a more consistent and compatible view on the topic. Three major databases were systematically searched for articles on related topics (Science Direct, IEEE, Google Scholar). Through a systematic study, nine research papers related to this topic are obtained. Results show that this topic’s work is still recent, and many methods have not yet been tested to figure out the best approach for classifying Arabic texts as cyberbullying. Also, the number of available datasets for cyberbullying classification in Arabic text is still limited compared to the English language. Therefore, generalizing the best method and dataset for classifying cyberbullying is so difficult. Furthermore, the majority of studies used Twitter for collecting the datasets. SVM is the most used classifier, Whereas CNN, is the most commonly used neural network for the cyberbullying classification.