Table 4.

Summary of the themes related to each research question.

Research Question	Theme	Description	Sources
RQ1: Current trends in cyberbullying detection for Arabic language and dialects	ML and DL Approaches	ML models (e.g., SVM, Naïve Bayes) and DL models (e.g., CNN, BERT) are common for cyberbullying detection, with ensemble methods improving accuracy.	Haidar et al. (2017); Alakrot et al. (2018); Alrashidi et al. (2023)
	Sentiment Analysis and Lexicon-Based Methods	Sentiment analysis and lexicon-based approaches capture emotional tones and harmful language, essential for handling Arabic’s diverse dialects.	AlHarbi et al. (2019); Farid and El-Tazi (2020)
	Handling Arabic Dialects and Complexity	Specialized datasets and models (e.g., AraBERT, multilingual BERT) address dialectal variability, enhancing model accuracy for Arabic.	Mubarak and Darwish (2019); AbdelHamid et al. (2022); Khezzar et al. (2023)
RQ2: Standards used for detecting cyberbullying based on its characteristics	Development of Cyberbullying Datasets	Creation of Arabic-specific datasets that include dialectical variations and cyberbullying characteristics, though issues like imbalanced datasets (few cyberbullying instances) impact model performance.	Bashir and Bouguessa (2021); Khairy et al. (2023); AbdelHamid et al. (2022)
	Evaluation Standards and Metrics	Precision, recall, F1-score, and accuracy are commonly used metrics, supplemented by specialized metrics tailored to Arabic-language characteristics to ensure reliable detection performance.	Haidar et al. (2017); Alakrot et al. (2021); Boulouard et al. (2022)
	Linguistic and Psychological Standards	Integration of linguistic and psychological insights, such as personality inference, allows a deeper understanding of user behavior, helping to identify cyberbullying based on more human-centered behavioral traits.	Elzayady et al. (2023); Omar et al. (2021); Shannaq et al. (2022)
	Contextual and Cultural Considerations	Incorporation of cultural sensitivity, including the use of dialect-specific language features, emojis, and contextual sentiment, provides a more nuanced and culturally accurate detection of offensive language.	AlHarbi et al. (2019); Farid and El-Tazi (2020); Khezzar et al. (2023)
RQ3: Future research directions for Arabic cyberbullying detection	Dialect-Specific Datasets and Multilingual Models	Expansion of dialect-specific datasets and multilingual models to enhance detection across Arabic dialects and improve cross-linguistic applicability.	Ali and Kurdy (2022); Rachidi et al. (2023); Shannaq et al. (2022)
	Advanced Feature Engineering and Hybrid Models	Development of hybrid models (e.g., CNN-LSTM-BERT) and advanced feature engineering, such as attention mechanisms and personality-based features, for richer context and improved detection accuracy.	Mouheb et al. (2019); Elzayady et al. (2023); Boulouard et al. (2022)
	Real-Time Detection and Privacy Considerations	Focus on real-time cyberbullying detection models for immediate response, with privacy-preserving techniques to ensure user data protection and ethical AI application.	Amer Hamzah and Dhannoon (2023); Omar et al. (2021); Kanan et al. (2021)
	Cross-Disciplinary Research	Integration of psychological, sociological, and linguistic insights for a more comprehensive understanding of the social and behavioral dynamics underlying Arabic cyberbullying.	Farid and El-Tazi (2020); Omar et al. (2021); Elzayady et al. (2023)