A Novel Approach to Hostility Detection: Employing GPT-2 for Text Classification

Social media platforms provide an easy way to share valuable information and convey personal opinions. Even though these platforms have many benefits, they have made aggressive posts more common. People post hate speech and insulting comments for their own pleasure or to help their political campaigns. The entire platform experience can be rendered hostile by the harassing effect of the hostile posts. Therefore, it is crucial to identify antagonistic posts in order to preserve social media hygiene. Languages with limited resources, such as Hindi, are particularly susceptible to this issue. In this study, we introduce methods for detecting hostile text in the Hindi language. The Constraint@AAAI 2021 Hindi hostility detection dataset is used to evaluate the proposed approaches. For this multi-label classification problem, we assess a variety of deep learning methods that are based on GPt-2. We demonstrate that the performance of models based on GPT-2 is optimal.