KCR: Resolving Long-Context Knowledge Conflicts via Reasoning in LLMs

2025-08-02

AI Summary

Large language models (LLMs) often suffer from reasoning confusion when resolving multi-source knowledge conflicts in long-context settings. To address this, we propose the Knowledge Conflict Resolution (KCR) framework—a novel approach that jointly models reasoning paths via textual and local knowledge graph representations and integrates reinforcement learning with a logic-consistency reward function. This reward explicitly guides the model to identify and prioritize semantically coherent, evidence-supported reasoning chains. Crucially, KCR enables LLMs to explicitly learn conflict detection and resolution across disparate contextual sources, moving beyond implicit pattern matching. Extensive experiments on multiple long-context knowledge conflict benchmarks demonstrate that KCR significantly improves conflict resolution accuracy across mainstream models—including Llama-3, Qwen2, and DeepSeek-V3—with an average gain of 12.7%. These results validate KCR’s effectiveness and generalizability in mitigating knowledge conflicts within extended contexts.

Technology Category

• Knowledge Representation and Reasoning: Knowledge Representation Languages
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Knowledge conflicts commonly arise across diverse sources, and their prevalence has increased with the advent of LLMs. When dealing with conflicts between multiple contexts, also known as emph{inter-context knowledge conflicts}, LLMs are often confused by lengthy and conflicting contexts. To address this challenge, we propose the Knowledge Conflict Reasoning (KCR) framework, which enhances the ability of LLMs to resolve conflicting knowledge. The key idea of KCR is to train backbone LLMs to establish a correct reasoning process by rewarding them for selecting and adhering to the context with stronger logical consistency when presented with conflicting contexts. Specifically, we first extract reasoning paths, represented by either text or local knowledge graphs, from the conflicting long contexts. Subsequently, we employ Reinforcement Learning to encourage the model to learn the paradigm of reasoning process that follows correct reasoning paths rather than the incorrect counterparts. This enables the backbone models to genuinely acquire the capability to resolve inter-context knowledge conflicts within long contexts. Experimental results demonstrate that our framework significantly improves the ability of various backbone models to resolve knowledge conflicts in long-context scenarios, yielding substantial performance gains.

Problem

• Resolving knowledge conflicts in lengthy LLM contexts
• Enhancing LLM reasoning for logical consistency
• Training models to select correct reasoning paths

Innovation

• Uses Knowledge Conflict Reasoning (KCR) framework
• Extracts reasoning paths from conflicting contexts
• Employs Reinforcement Learning for correct reasoning