Title: IDO: Incongruity-aware Distribution Optimization for Multimodal Fake News Detection

Abstract

The primary objective of multimodal fake news detection is to verify the authenticity of news content. While current methodologies predominantly emphasize cross-modal consistency, they frequently overlook the explicit modeling of semantic incongruity, a hallmark of deceptive multimaterials. Misinformation is often characterized by semantic discrepancies that contradict factual reality. To mitigate these limitations, we introduce Incongruity-aware Distribution Optimization (IDO), a framework designed to enhance detection efficacy by addressing both factual and modality incongruities. In the context of factual incongruity, our approach employs a channel-wise reweighting strategy to generate semantically discriminative embeddings, alongside the use of Gaussian distributions to capture uncertain correlations arising from factual inconsistencies. To handle modality incongruity, we implement incongruity contrastive learning to effectively acquire cross-modal semantic features. Our experimental results confirm that IDO delivers state-of-the-art performance.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC