Sustainable signals: a heterogeneous graph neural framework for fake news detection

Digital technology has increased the spread of fake news, leading to misconceptions, misunderstandings, and economic challenges. Researchers have developed automated techniques to identify false information using various data features, driven by advancements in AI. Most algorithms focus on signals from the news itself and its context, often ignoring user preferences. According to confirmation bias theory, individuals are more likely to spread false information that aligns with their beliefs. Users’ historical and social activities, such as their postings, can help identify fake news and inform their news choices. However, there is limited research on incorporating user preferences in fake news detection. This study introduces a framework based on Graph Neural Networks (GNNs) and natural language models to capture signals from both graph and content perspectives, considering user preferences. We chose GNNs for their ability to model complex relationships in graph-structured data. Specifically, we used the Graph Attention Network due to its ability to weigh the importance of different nodes, enhancing the capture of relevant signals. The framework integrates user preferences by analyzing social activities and news choices. Experimental results on a real-world dataset show our model achieves an accuracy of 98%. Outperforming models that do even consider user preferences. These findings highlight the potential of leveraging user preferences to enhance fake news detection, offering a more robust approach to tackling information pollution.