Inconsistent Matters: A Knowledge-guided Dual-consistency Network for Multi-modal Rumor Detection

Rumor spreaders are increasingly utilizing multimedia content to attract the attention and trust of news consumers. Though quite a few rumor detection models have exploited multi-modal data, they seldom consider inconsistent semantics between images texts, rarely spot inconsistency among post contents background knowledge. In addition, commonly assume completeness multiple modalities thus incapable handling handle missing in real-life scenarios. Motivated by intuition that rumors social media more likely semantics, novel Knowledge-guided Dual-consistency Network is proposed detect with contents. It uses two consistency subnetworks capture at cross-modal level content-knowledge simultaneously. also enables robust representation learning under different visual modality conditions, using special token discriminate posts without modality. Extensive experiments on three public real-world datasets demonstrate our framework can outperform state-of-the-art baselines both complete incomplete conditions. Our codes available https://github.com/MengzSun/KDCN.