Bert-Based Transformer Model for Hate Speech Detection
Puvvadi Harsha Vardhan, Hari Prasad, Bandi Vishnu Swaroop, Busa Thanuj Sathwik Reddy, Manju Venugopalan · 2024
Hate speech recognition has emerged as a key problem in social media and should be addressed using multiple models. Machine learning models like CNN and LSTM have been used recently in creating successful hate-speech detection models that have been developed in the recent past but they have not proved to be highly discriminative. The use of BERT or any other transformer model will give increased performance owing to the ability of these models to grasp context. Other potential improvements include using new feature engineering techniques and ensemble methods in conjunction with these models. The datasetused for experimentation in the proposed work is annotated considering three categories hate, offensive, or normal. The proposed methodology has incorporated a data-balancing mechanism and experimented with a range of classifier models that belong to ML and DL models and the best results were reported for Bert based transformer model which is F1-score of 0.76.