The increased use of social media, hate speech poses significant challenges, especially in multilingual regions. This paper evaluates contemporary deep learning models—MuRIL, BERT, RoBERTa, IndicBERT, and XLM-RoBERTa—for Hindi hate speech classification using a dataset of 8,192 multicategorical posts. MuRIL and XLM-RoBERTa achieved F1-scores of 0.955 and 0.945, excelling in code-mixed Hindi-English contexts. The study underscores the need for larger, diverse datasets to enhance model generalization and discusses emerging trends in multilingual hate speech detection.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Enhancing Hindi Hate Speech Detection: Multilingual & Multimodal Deep Learning Solutions

  • Rachna Narula,
  • Udit Hasija,
  • Vijay Kumar,
  • Nitasha Rathore

摘要

The increased use of social media, hate speech poses significant challenges, especially in multilingual regions. This paper evaluates contemporary deep learning models—MuRIL, BERT, RoBERTa, IndicBERT, and XLM-RoBERTa—for Hindi hate speech classification using a dataset of 8,192 multicategorical posts. MuRIL and XLM-RoBERTa achieved F1-scores of 0.955 and 0.945, excelling in code-mixed Hindi-English contexts. The study underscores the need for larger, diverse datasets to enhance model generalization and discusses emerging trends in multilingual hate speech detection.