Exploring Arabic Hate Speech Detection: Dialect Variations and Multimodal Approaches

Abstract

With the growing use of social networks and online platforms, hate speech and racist content detection is becoming increasingly challenging, especially in the Arab world, which is characterized by a significant diversity of complex dialects. This article highlights the importance of implementing detection systems capable of identifying hate speech, based on a critical analysis of scientific literature published between 2020 and 2025. This paper highlights the diversity of Arabic dialects and their linguistic complexity, which make them difficult to analyze using a single model. As a result, the utility of using multimodal approaches for detecting hate speech is emphasized. The article presents this approach and evaluates the effectiveness of several artificial intelligence-based methods, implementing methodological innovations and identifying gaps in current research, as well as addressing key challenges such as the limited consideration of Maghrebi dialects, the lack of datasets for these dialects, especially Moroccan Darija.