[论文解读] ExDR: Explanation-driven Dynamic Retrieval Enhancement for Multimodal Fake News Detection

Guoxuan Ding, Yuqing Li+4 more|arXiv (Cornell University)|Jan 22, 2026

Misinformation and Its Impacts被引用 0

一句话总结

ExDR 利用模型生成的解释动态触发检索并检索对比性、实体丰富的证据，从而提升多模态假新闻检测。

ABSTRACT

The rapid spread of multimodal fake news poses a serious societal threat, as its evolving nature and reliance on timely factual details challenge existing detection methods. Dynamic Retrieval-Augmented Generation provides a promising solution by triggering keyword-based retrieval and incorporating external knowledge, thus enabling both efficient and accurate evidence selection. However, it still faces challenges in addressing issues such as redundant retrieval, coarse similarity, and irrelevant evidence when applied to deceptive content. In this paper, we propose ExDR, an Explanation-driven Dynamic Retrieval-Augmented Generation framework for Multimodal Fake News Detection. Our framework systematically leverages model-generated explanations in both the retrieval triggering and evidence retrieval modules. It assesses triggering confidence from three complementary dimensions, constructs entity-aware indices by fusing deceptive entities, and retrieves contrastive evidence based on deception-specific features to challenge the initial claim and enhance the final prediction. Experiments on two benchmark datasets, AMG and MR2, demonstrate that ExDR consistently outperforms previous methods in retrieval triggering accuracy, retrieval quality, and overall detection performance, highlighting its effectiveness and generalization capability.

研究动机与目标

• 在知识不断演化和内容存在欺骗性的背景下，推动多模态假新闻的鲁棒检测。
• 开发面向多模态数据的动态 Retrieval-Augmented Generation 框架。
• 利用模型生成的解释来引导检索触发与证据检索。
• 通过实体感知的索引与对比证据提高检索效率与检测准确性。

提出的方法

• 提出基于三个以模型为中心的置信维度的检索触发：标签级不确定性、令牌级支持、句子级置信度。
• 通过两阶段混合检索（全局探索与局部精细化）来优化检索阈值。
• 通过融合视觉、文本和解释派生的实体特征，构建实体增强的多模态混合索引。
• 从解释中推断细粒度的欺骗标签，并在这些标签的指导下执行对比性证据检索（正负样本）。
• 在融合特征上构建 FAISS 索引，以实现高效相似性搜索并检索定向证据。
• 用两项新指标（Retrieval Identification Rate 与 Retrieval Efficiency）进行评估，并在 AMG 与 MR2 数据集上与基线方法进行对比。

Figure 1 . Overview of our proposed ExDR framework. ExDR consists of two main components: (1) a retrieval triggering module that dynamically determines whether retrieval is necessary based on response analysis, and (2) an evidence retrieval module that retrieves targeted evidence, including both pos

实验结果

研究问题

• RQ1 在多模态假新闻检测中何时触发检索？
• RQ2 如何利用解释与置信信号构建有效的检索查询？
• RQ3 应检索哪种类型的欺骗样本以最佳辅助决策？
• RQ4 实体增强索引是否提升检索质量与证据相关性？
• RQ5 对比性、欺骗感知的证据检索是否能提升最终检测性能？

主要发现

• ExDR 触发在 AMG 与 MR2 上对比基线具有更高的 Retrieval Identification Rate 与 Retrieval Efficiency。
• 实体增强的多模态索引为证据检索提供了更具判别力的线索，提升准确性与鲁棒性。
• 对比性证据检索（正样本与负样本）进一步提升性能，特别在普通 LVLM 与跨域 MR2 上表现突出。
• 端到端的 ExDR 在 AMG 上的 ACC 与 F1 指标持续优于强基线 MFND（如 MGCA）。
• 消融研究表明所有组件（多层级置信触发、实体索引、基于标签的信息驱动对比检索）对性能有显著贡献。

Figure 2 . Case study illustrating the effectiveness of ExDR across three representative scenarios.