DRAGged into Conflicts: Detecting and Addressing Conflicting Sources in Search-Augmented LLMs

2025-06-10

AI Summary

In real-world RAG systems, multi-source retrieval often yields conflicting knowledge, yet current LLMs lack systematic capabilities to identify and reconcile such conflicts. Method: We propose the first knowledge conflict taxonomy tailored to practical RAG scenarios and introduce CONFLICTS—a high-quality, expert-annotated benchmark enabling fine-grained evaluation of conflict-aware response generation. Through controlled RAG experiments and a novel Chain-of-Conflict prompting strategy, we characterize prevalent LLM failure modes in conflict resolution—including misclassification, omission, and arbitrary preference. Contribution/Results: Experimental results demonstrate that explicitly prompting models to recognize, articulate, and reflect upon conflicts significantly improves response accuracy and coherence. This work establishes a theoretical foundation for conflict-aware RAG, provides a standardized evaluation benchmark, and delivers a reproducible methodological framework for mitigating knowledge conflicts in retrieval-augmented generation.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Knowledge Representation and Reasoning: Knowledge Acquisition
• Multiagent Systems: Agreement, Argumentation & Negotiation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Retrieval Augmented Generation (RAG) is a commonly used approach for enhancing large language models (LLMs) with relevant and up-to-date information. However, the retrieved sources can often contain conflicting information and it remains unclear how models should address such discrepancies. In this work, we first propose a novel taxonomy of knowledge conflict types in RAG, along with the desired model behavior for each type. We then introduce CONFLICTS, a high-quality benchmark with expert annotations of conflict types in a realistic RAG setting. CONFLICTS is the first benchmark that enables tracking progress on how models address a wide range of knowledge conflicts. We conduct extensive experiments on this benchmark, showing that LLMs often struggle to appropriately resolve conflicts between sources. While prompting LLMs to explicitly reason about the potential conflict in the retrieved documents significantly improves the quality and appropriateness of their responses, substantial room for improvement in future research remains.

Problem

• Detecting conflicting sources in search-augmented LLMs
• Addressing discrepancies in retrieved information for RAG
• Benchmarking model performance on resolving knowledge conflicts

Innovation

• Novel taxonomy for RAG knowledge conflicts
• CONFLICTS benchmark with expert annotations
• Prompting LLMs to reason about conflicts