Identification of social network automated hate speech using GLTR with BERT and GPT-2 : A novel approach

Abstract

Automated hate speech on social media is a serious issue for online safety and wellbeing. A promising strategy to address the issue is the identification of artificial hate speech using machine learning techniques. The authors present research on the detection of automated hate speech on social networks with the use of GLTR (Giant Language model Test Room) along with BERT (Bidirectional Encoder Representations from Transformers) as well as GPT-2 (Generative Pretrained Transformer 2) on the four widely used benchmark datasets (ETHOS, Founta, Waseem, and Davidson). The performance of the proposed approach was evaluated by comparing F1 scores, precision, recall, and accuracy with other state-of-the-art machine algorithms. The results show that the proposed approach out performs other algorithms on all four datasets. The results indicate that the GLTR approach with BERT and GPT-2 models is a promising method for automated hate speech detection on social networks.