Toward Detection of Arabic Cyberbullying on Online Social Networks using Arabic BERT Models
Meshari Essa AlFarah, Ibrahim Kamel, Zaher Al Aghbari · 2023
Cyberbullying is one of the serious threats on social networks particularly toward children and teenagers. Cyberbullying can cause many harmful consequences towards victims such as anxiety, depression and even suicide. This research studies cyberbullying detection in Arabic language on online social networks using deep learning techniques to automatically detect and quarantine the cyberbullying messages to safe children from getting exposed to harmful cyberbullying content. In this paper, real Arabic dataset is collected from YouTube and Twitter that is annotated manually to improve the quality of the data. All experiments went through three rounds of trials to make sure that the results are consistent. Many evaluation metrics were used to evaluate the performance of the classifiers such as macro F1 score and AUC. The best model achieves 84.58% and 85.94% in F1 score and AUC respectively.