Explainable Detection of Fake News and Cyberbullying on Social Media

While social media had ubiquitously penetrate into into people’s daily life, where allows interactions between people, user-generated text data not only enables novel applications, but also provides user digital footprints for us to analyze a variety of human behaviors. In this talk, we will share two of our recent studies on combating anti-social behaviors: detecting fake news and identifying cyberbullying behaviors on social media. We will reveal three important insights. First, it is possible to predict anti-social behaviors without social network information. Second, graph neural networks (GNN) is effective in improving the performance of such two tasks. Third, our models can provide model explainability to understand the language use of anti-social behaviors. In the end of this talk, we will point out future directions on fighting with fake news and cyberbullying in social media.