Graph-aware multi-feature interacting network for explainable rumor detection on social network

Abstract

At present, rumors are growing wantonly with the convenience and influence of social media, becoming a problem that may severely impact social stability and development. The rumor is not an objective judgment but a process of multi-dimensional subjective value superposition and a collective transaction of people’s thoughts on social networks. How to fully mine the critical features of rumor detection from the complex information of social networks is a challenge to the existing rumor detection models. Therefore, we present the explainable model GMIN (Graph-aware Multi-feature Interacting Network), aiming to fully exploit the multifaceted features of social networks by deeply analyzing the mechanism of rumor propagation and the nature of social networks. GMIN incorporates four modules to capture: the characteristics of people receiving and spreading information, the interactions and latent associations in social networks, the mechanisms of rumor propagation and diffusion, and the collaborations between various features. It is worth noting that GMIN is interpretable in the results and the model’s components. The experimental results on three real-world datasets demonstrate the validity of our proposed model and show excellent capabilities in early rumor detection and the interpretability of the model.