Team curie at HSD-2Lang 2024:hate speech detection in Turkish and Arabic tweets using BERT-based models

Abstract

This study focuses on hate speech detection in Turkish and Arabic tweets using advanced BERT-based models. Performance metrics demonstrate the models' effectiveness, with the Turkish variant achieving a 71.8% F1 score and the Arabic model a 76.9% F1 score, ranking them fourth and third, respectively, in a competitive leaderboard. Performance enhancements were realized through targeted preprocessing, including emoji translation and user mention exclusion, and thoughtful data balancing approaches. Future directions include refining model accuracy and broadening language support. Our reproducible approach and detailed findings are accessible on GitHub.