Detecting Offensive Tamil Texts Using Machine Learning And Multilingual Transformer Models

Nowadays, social media has permitted an exponential increase in the circulation of hostile and toxic content, which has resulted in an increase in the number of people exposed to it. Many members of the Natural Language Processing community have recently expressed an interest in automated identification of such harmful content as hate speech, provocative language, and abusive language as a means of addressing this problem. Machine learning and multilingual transformer models are used in this study to automatically identify Tamil language comments as either offensive or not offensive messages. The dataset is collected from YouTube and Kaggle. BERT tops the competition when it comes to offensive language identification models, with an accuracy of 82% compared to the others.