A Multifaceted Reasoning Network for Explainable Fake News Detection

Abstract

Fake news detection involves developing techniques to identify and flag misleading or false information disseminated through media sources. Current efforts often use limited information for categorization, lacking comprehensive data integration and explanation of results. Additionally, the substantial noise generated by multi-source data presents extra challenges to fake news detection. To address these problems, we propose a novel Multifaceted Reasoning Network for Explainable Fake News Detection (MRE-FND). This model constructs two heterogeneous graphs to learn about social network information and news content knowledge, including news content, social networks, knowledge graphs, and external news data. Utilizing graph information bottleneck theory, it eliminates noise from multifaceted data and extracts key information for fake news detection. An interpretable reasoning module is designed to provide clear explanations for the classification results. Our proposition undergoes extensive evaluation on three popular datasets, Politifact, Gossipcop and Pheme, which consist of 495, 15707 and 2189 news, respectively. Our model achieved state-of-the-art results across all metrics on three datasets. Specifically, our model achieved accuracy rates of 92.9%, 83.4% and 84.7% on the Politifact, Gossipcop and Pheme datasets, respectively, demonstrating improvements of 2.0, 0.8 and 1.1 percentage points over the baseline, thus establishing the superiority of our model. Further analysis indicates that our model can effectively handle redundant information in multi-faceted data, enhancing the performance of fake news detection while also providing multifaceted explanations for the classification results.