Multi View Learning-Based Graph Neural Network for Rumor Detection

While existing graph neural network methods have integrated users' social behaviors and text content to some extent for rumor detection, they primarily suffer from two limitations. First, the propagation patterns of rumors are often described by static graphs, which presuppose that the structure of the rumor propagation network is fixed before the algorithm learns, overlooking the fact that rumor dissemination is a dynamic process. Second, current methods tend to neglect the in-depth exploration of complex interactions and dynamic changes between different information dimensions. Some approaches merely integrate personal user information as model inputs, others rely solely on time series analysis to track the changes in rumors, or fail to effectively differentiate the impact of various user features on rumor spread. response to these issues, this study introduces a new model framework named Multi-View Graph Neural Network (MV-GNN), which utilizes different perspectives to represent rumors, capturing more comprehensive information. We first abstract the process involved in rumor propagation into three perspectives: the user profile perspective, the user comment perspective, and the rumor diffusion perspective. Then, we integrate these different views to predict the credibility of given information. On two real-world datasets, MV-GNN achieved the best detection performance, demonstrating the efficacy of the model.