Enhancing Hate Speech Detection in Tamil Code-Mix Content: A Deep Learning Approach with Multilingual Embeddings

The increasing amount of hate speech and language which are inappropriate on platforms of social media has emerged as a serious issue, prompting measures to counter it. This study examines hate speech and languages that are objectionable in a Tamil Code-Mix dataset, where Tamil content is provided in English. It assesses the performance of various deep learning classifiers that use different multilingual transformer embedding models, which includes many models such as DistilBERT, IndicBERT, XLM-RoBERTa, LASER, MuRIL, and BERT, in this domain. Among the models tested, an ensemble stacked model integrating Long Short-Term Memory and Gated Recurrent Units with XLM-RoBERTa embeddings had the greatest accuracy (0.76) and weighted F1-score (0.72). This exceptional performance underscores the potential of advanced ensemble models for distinguishing hate speech in code-mixed languages. The study's findings are notable for several reasons. They enhance understanding of hate speech in Tamil, particularly in its code-mixed variant, and offer insights into the application of advanced deep learning methods and multilingual embeddings for detecting harmful content. By improving accuracy and reliability in hate speech detection, this research contributes to fostering safer and more inclusive online environments for Tamil-speaking communities. In summary, the research advances the domain of hate speech identification and emphasizes the importance of technological solutions in creating more safer and very inclusive online spaces.