Multi-modal Rumor Detection via Knowledge-Aware Heterogeneous Graph Convolutional Networks

Boqun Li, Zhong Qian, Peifeng Li, Qiaoming Zhu

Abstract: With the rapid growth of the number of social media users, a variety of unverified information inevitably spreads on the social platform, which leads to the diffusion of rumors. Although some methods are explored on multi-modal data, they seldom take into account the hidden knowledge behind the text and image, and ignore the widely dispersed structure on multi-modal data in the rumor detection field. To solve the above issues, we propose a novel Multi-Modal Rumor detection model via Knowledge-aware Heterogeneous Graph Convolutional Networks, i.e., M $$^3$$ KHG, which can model a post as a propagation graph, capture the interactive semantic information of image and text at the cross-modal level, and highlight suspicious signals according to the correlation between text-image knowledge in a unified framework. Finally, the “knowledgeable” feature generated by the propagation graph is assigned to debunk rumors. Experimental results on three popular datasets show that our model M $$^3$$ KHG is superior to the state-of-the-art baselines.