DLRG@DravidianLangTech 2025: Multimodal Hate Speech Detection in Dravidian Languages
R. Rajalakshmi, Ramesh Kannan, Meetesh Saini, Bitan Mallik · 2025
Social media is a powerful communication tool and rich in diverse content requiring innovative approaches to understand nuances of the languages.Addressing challenges like hate speech necessitates multimodal analysis that integrates textual, and other cues to capture its context and intent effectively.This paper proposes a multi-modal hate speech detection system in Tamil, which uses textual and audio features for classification.Our proposed system uses a fine-tuned Indic-BERT with Whisper as a multimodal approach for hate speech detection.The fine-tuned Indic-BERT model with Whisper achieved an F1 score of 0.25 on Multimodal based approach.Our proposed approach ranked at 10th position in the shared task on Multimodal Hate Speech Detection in Dravidian languages at the NAACL 2025 Workshop DravidianLangTech.