Multi-view learning with distinguishable feature fusion for rumor detection

Researchers, enterprises, and governments have made great efforts to detect misinformation promptly accurately. Traditional solutions either examine complicated hand-crafted features or rely heavily on the constructed credibility networks extract useful indicators for discerning false information. However, such approaches require insightful domain expert knowledge intensive feature engineering that are often non-generalizable. Recent advances in deep learning techniques spurred high-level representations from textual image content discovering diffusion patterns with various neural networks. Despite progress by these methods, they still face problem of overdependence fail discriminate against influence each user involved process rumor spreading. Different user-aspect information plays different roles stages diffusion, effectively aspect, aggregate learned into a unique representation, which has not been well investigated. To address limitations, we propose novel model, UMLARD (User-aspect Multi-view Learning Attention Rumor Detection), learn representation views users who engaged spreading tweet, fuse through distinguishable fusion mechanism. Finally, concatenate form feed it fully connected layer predict label rumors. Our experiments conducted real-world datasets demonstrate significantly improves detection performance compared state-of-the-art baselines. It also allows explainability model behavior predicted results. • A multi-view detection. hierarchical way represent diffusion. mechanism unify knowledge. Extensive prove effectiveness our model.