Enhancing Hate Speech Detection with Integrated Content-Based and Stylistic Features

Abstract

The proliferation of harmful and unpleasant speech on community medium platforms has underscored the need for effective hate speech detection. While recent efforts have focused on refining pre-trained models, this study takes a novel approach by emphasizing the integration of content-based and stylistic features. Stylistic features, in particular, play a critical role in hatred speech detection. By capturing unique linguistic patterns and characteristics indicative of hateful or offensive language—beyond explicit content—these features enhance the discriminatory power of detection systems. In this research, exploration of combined utilization of SVM, XGBoost, and Random Forest algorithms on a comprehensive dataset. The results surpass existing methodologies, contributing to more effective identification and mitigation of problematic content online.