Flexible Framework to Provide Explainability for Fake News Detection Methods on Social Media

The spread of fake news on social media has become a global problem that is attracting attention to fake news detection methods that use machine learning models. Recently, social context-based detection methods have been developed to learn users’ endogenous preferences from propagation patterns using graph neural networks, focusing on the way that news spreads on social media. However, the computational process of these methods is a black box, and the evidence for detection is unknown. Explanations of the rationale are very important to provide insight for understanding the model and obtaining trust. In this paper, we propose a framework that can add explainability to existing social context-based detection methods in a flexible manner. We quantitatively evaluated the effectiveness of the proposed explanations through experiments on a real-world dataset extracted from Twitter.