Resolving Conflicting Evidence in Automated Fact-Checking: A Study on Retrieval-Augmented LLMs

2025-05-23

AI Summary

Retrieval-augmented generation (RAG) models exhibit insufficient robustness in factual verification when confronted with conflicting evidence from multiple sources—particularly due to disparities in source credibility (e.g., media trustworthiness). Method: We introduce CONFACT, the first benchmark explicitly designed for factual verification under conflicting evidence; systematically identify critical failures of RAG in credibility-aware conflict resolution; and propose a novel cross-stage framework integrating media context—comprising credibility-aware retrieval ranking, source-aware prompt engineering, and multi-source conflict modeling and resolution. Results: Experiments demonstrate substantial improvements in RAG’s verification accuracy under conflict, confirming that explicit source credibility modeling yields fundamental gains in generation reliability. CONFACT establishes a reproducible evaluation benchmark and provides methodological foundations for trustworthy RAG.

Technology Category

Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)Knowledge Representation and Reasoning: Reasoning with BeliefsReasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty

Application Category

Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAGUser Modeling, Personalization and Recommendation: Fairness-aware retrieval and rankingWeb Mining and Content Analysis: Web data provenance, reliability, and authenticity

Abstract

Large Language Models (LLMs) augmented with retrieval mechanisms have demonstrated significant potential in fact-checking tasks by integrating external knowledge. However, their reliability decreases when confronted with conflicting evidence from sources of varying credibility. This paper presents the first systematic evaluation of Retrieval-Augmented Generation (RAG) models for fact-checking in the presence of conflicting evidence. To support this study, we introduce extbf{CONFACT} ( extbf{Con}flicting Evidence for extbf{Fact}-Checking) (Dataset available at https://github.com/zoeyyes/CONFACT), a novel dataset comprising questions paired with conflicting information from various sources. Extensive experiments reveal critical vulnerabilities in state-of-the-art RAG methods, particularly in resolving conflicts stemming from differences in media source credibility. To address these challenges, we investigate strategies to integrate media background information into both the retrieval and generation stages. Our results show that effectively incorporating source credibility significantly enhances the ability of RAG models to resolve conflicting evidence and improve fact-checking performance.

Problem

• Evaluating RAG models for fact-checking with conflicting evidence
• Addressing reliability issues from varying source credibility
• Enhancing RAG performance by integrating source credibility

Innovation

• Retrieval-Augmented LLMs for fact-checking
• Introducing CONFACT dataset with conflicts
• Integrating source credibility into RAG models