Multi-modal Rumor Detection on Modality Alignment and Multi-perspective Structures

Boqun Li, Zhong Qian, Peifeng Li, Qiaoming Zhu

Abstract: Due to the rapid spread of rumors on social media and their negative impact on society, real-time rumor detection is of utmost importance. Although some rumor detection methods have applied the structure of temporal or graphic information, they do not consider multiple structures to obtain better representation. Besides, since the authors maybe only post texts in real-world social scenarios, image modalities become inaccessible in multi-modal rumor detection. To solve the above issues, we propose a Multi-Modal rumor detection model on Modality Alignment and multi-Perspective Structures (M3APS). The model uses the image generation method to fill the inaccessible image modalities in the multi-modal heterogeneous node pair and fuses the node pair to obtain multi-modal features. Then, the “debunkers” which are extracted from the perspective of temporal structure and graphic structure query the events described in the source tweet, respectively. Experimental results on three popular datasets show that our model M3APS is superior to the state-of-the-art baselines.