Transformer-based models for hate speech classification (ICIASC)

In order to improve hate speech identification, this research study investigates the use of text categorization and natural language processing algorithms. To increase the accuracy of hate speech classifiers, the study uses deep learning and machine learning (ML) models, including transformer models like BERT, RoBERTa, and DistilBERT. The article shows how well these algorithms recognize hate speech by a thorough empirical investigation on three different datasets: Data-ICWSM, Data-ALW2, Data-OLID. The BERT models show a notable improvement in macro and weighted F1-score performance as compared to conventional baselines. The work also uses sampling strategies during training to solve the issue of uneven class distributions in the datasets.

https://pubs.aip.org/aip/acp/article/3072/1/020017/3277781/Transformer-based-models-for-hate-speech