Fine-Refine: Iterative Fine-grained Refinement for Mitigating Dialogue Hallucination

2026-02-17

AI Summary

This work addresses the prevalence of factual hallucinations in large language models during dialogue, which undermines system credibility. To mitigate this issue, the authors propose a fine-grained iterative refinement framework that moves beyond conventional response-level correction. The approach decomposes model responses into atomic factual units, verifies each unit against retrieved external knowledge, and iteratively refines the output while preserving fluency through perplexity-based evaluation. Evaluated on the HybriDialogue and OpendialKG datasets, the method significantly improves factual consistency—achieving gains of up to 7.63 points—while incurring only minimal degradation in conversational quality, thereby effectively balancing accuracy and naturalness.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• 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

The tendency for hallucination in current large language models (LLMs) negatively impacts dialogue systems. Such hallucinations produce factually incorrect responses that may mislead users and undermine system trust. Existing refinement methods for dialogue systems typically operate at the response level, overlooking the fact that a single response may contain multiple verifiable or unverifiable facts. To address this gap, we propose Fine-Refine, a fine-grained refinement framework that decomposes responses into atomic units, verifies each unit using external knowledge, assesses fluency via perplexity, and iteratively corrects granular errors. We evaluate factuality across the HybriDialogue and OpendialKG datasets in terms of factual accuracy (fact score) and coverage (Not Enough Information Proportion), and experiments show that Fine-Refine substantially improves factuality, achieving up to a 7.63-point gain in dialogue fact score, with a small trade-off in dialogue quality.

Problem

• dialogue hallucination
• large language models
• factuality
• response refinement
• atomic fact verification

Innovation

• fine-grained refinement
• dialogue hallucination
• fact verification
• iterative correction
• atomic fact decomposition