Gated Feature Fusion based Multi‑view Graph Attention for Social Rumor Classification

The widespread use of the Internet and the rapid development of social media have greatly facilitated people’s lives, but they have also significantly promoted the spread of rumors, even causing social unrest. Therefore, rumor detection has become a global public challenge. Although traditional machine learning methods have certain practicality, they often require manual extraction of data features and focus only on text analysis, ignoring the increasingly complex structure of rumors. Deep learning models can further effectively extract rumor features, but they ignore the importance of different features. In recent years, graph neural networks have shown great potential in rumor detection tasks, but existing models still have significant limitations in capturing long-range dependencies and integrating heterogeneous information. To address these issues, we propose a multi-view attention network named GFF-MGA based on the gated feature fusion mechanism. GFF-MGA simultaneously constructs two topological views: the tweet propagation subgraph and the user-tweet interaction subgraph. The tweet propagation subgraph is used to model the forwarding, reply, and propagation links between tweets, reflecting the rumor diffusion path. The user-tweet subgraph describes the publishing and forwarding behaviors between users and tweets. These two subgraphs can respectively capture content propagation features and user behavior features, compensating for the representation deficiencies caused by the lack of information in a single view topology. We conducted experiments on the Twitter15 and Twitter16 datasets, and the results showed that our method outperformed the current baseline models, achieving an accuracy improvement of 1.4% and 0.2% on the two datasets.