Arabic Hate Speech Identification by Enriching MARBERT Model with Hybrid Features

Abstract

The propagation of hate speech has grown more particularly apparent on social media sites as their usage for communication and unrestricted thought has increased on a worldwide scale. However, this could cause disagreement and antagonism among users, creating an unattractive online environment. Countries, companies, and academic institutions have all invested heavily in finding an effective solution to this challenge. There has been less study done in Arabic compared to other languages to develop automated systems for recognizing hate speech. Additionally, Arabic research on the correlation between personality traits and hate speech still remains rather limited. In this paper, we propose a novel method for enriching MARBERT model that incorporates static word embedding (AraVec 2.0) and personality trait features for Arabic hate speech detection. The experimental results indicate that the suggested methodology exceeds in terms of macro-F1 score by achieving 86.4% compared to previous research reported in the literature.