Multilingual Hate Speech Identification in Hindi, English and Marathi with Deep Learning Techniques

Abstract

Given the extensive adoption of the social media platforms, detecting hate speech and offensive language is essential for maintaining community unity and managing digital content. While existing research primarily concentrates on single-language training within a multilingual context, our study introduces a model trained across multiple languages. Our hybrid methodology incorporates mBERT and MuRILBERT models customized for English, Hindi, and Marathi. The model exhibited exceptional performance in English and demonstrated moderate efficacy in Hindi and Marathi.