Abusive Content Detection in Arabic Tweets Using Multi-Task Learning and Transformer-Based Models

Different social media platforms have become increasingly popular in the Arab world recent years. The increasing use of media, however, has also led to emergence a new challenge form abusive content, including hate speech, offensive language, and language. Existing research work focuses on automatic content detection as binary classification problem. In addition, existing task surrounding Arabic fails tackle dialect-specific phenomenon. Consequently, this two important issues task. study, we used multi-aspect annotation schema problem countries, based multi-class dialectal (DA)-specific More precisely, includes five attributes: directness, hostility, target, group, annotator. We specifically developed framework automatically detecting Twitter using natural language processing (NLP) techniques. different models machine learning (ML), deep (DL), pretrained (LMs) dataset. investigate impact other approaches, such multi-task (MTL), four MTL built top DA model (called MARBERT) trained Our LMs enhanced performance compared DL mentioned literature.