Investigating the Predominance of Large Language Models in Low-Resource Bangla Language over Transformer Models for Hate Speech Detection: A Comparative Analysis (MDPI)

There are 35,197 comments in the training set, 7542 in the validation set, and 7542 in the test set of the BD-SHS dataset. Personal hatred (629), political hate (1771), religious hate (502), geographical hate (1179), and gender abusive hate (316) are among the hate categories included in the Bengali Hate Speech Dataset v1.0 and v2.0. There are 7,500 hateful and 7,500 non-hateful comments in the Bengali Hate Dataset. GPT-3.5 Turbo’s accuracy rates of 97.33%, 98.42%, and 98.53% were remarkable. Gemini 1.5 Pro, on the other hand, performed worse on every dataset. In particular, GPT-3.5 Turbo outperformed Gemini 1.5 Pro with noticeably more precision. These results demonstrate a 6.28% improvement in accuracy above 92.25% obtained with conventional approaches.

https://www.mdpi.com/2227-7390/12/23/3687