Exploring the Effectiveness of Transformer-Based Models and ML Algorithms for Bengali Hate Speech Classification

The number of internet users is growing every day. Social media is becoming more popular because the internet is accessible almost everywhere. People use social media to share their opinions and sometimes post offensive thoughts, known as hate speech, which can cause problems in society. There are many ways to detect hate speech in different languages, but there are few methods for Bengali comments and posts. In this study, we aim to identify various types of hate speech in Bengali social media posts and comments using transformer-based models. The transformer-based model 3cube-pune/bengali-bert achieved an accuracy of 75.99% and an f'1-score of 76.07%, which is higher than the other models we considered.