Department of Computer Engineering

The increasing amount of toxic contents is a major problem on social media. Due to isolated and closed mindsets, people tend to behave in mostly two kinds of toxic behavior: (i) Using hate speech towards people from different backgrounds and communities, and (ii) Spreading misinformation for different purposes including political campaigns and economic benefits. In this talk, I present some of the challenges and opportunities in automatic detection of hate speech and misinformation by natural language processing. For this purpose, I provide the experimental results and analyses from my recent studies that discuss several challenges including the lack of data resources, the characteristics of social media text, the adaptation of large language models, and the dependency on external knowledge resources. The natural language processing perspective also offers opportunities to investigate the generalization capability of large language models and the effect of bias in such models.