FactSelfCheck: Fact-Level Black-Box Hallucination Detection for LLMs

2025-03-21

AI Summary

Addressing the pervasive hallucination problem in large language model (LLM) generation, existing approaches are largely confined to sentence- or paragraph-level detection, lacking fine-grained factual localization. This paper proposes the first zero-shot, black-box, external-knowledge-free fact-level hallucination detection method: it parses input text into knowledge graph triples and quantifies the hallucination probability of each triple via consistency analysis across multiple LLM sampling responses. Without requiring training data or external knowledge bases, our method achieves detection accuracy comparable to state-of-the-art sampling-based baselines while significantly improving hallucination correction—boosting factual correctness by 35%, substantially outperforming sentence-level methods (which yield only +8%). The core contribution is the first framework enabling interpretable, precisely localizable, and high-accuracy quantification and detection of hallucinations at the atomic fact level.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Large Language Models (LLMs) frequently generate hallucinated content, posing significant challenges for applications where factuality is crucial. While existing hallucination detection methods typically operate at the sentence level or passage level, we propose FactSelfCheck, a novel black-box sampling-based method that enables fine-grained fact-level detection. Our approach represents text as knowledge graphs consisting of facts in the form of triples. Through analyzing factual consistency across multiple LLM responses, we compute fine-grained hallucination scores without requiring external resources or training data. Our evaluation demonstrates that FactSelfCheck performs competitively with leading sampling-based methods while providing more detailed insights. Most notably, our fact-level approach significantly improves hallucination correction, achieving a 35% increase in factual content compared to the baseline, while sentence-level SelfCheckGPT yields only an 8% improvement. The granular nature of our detection enables more precise identification and correction of hallucinated content.

Problem

• Detects fine-grained fact-level hallucinations in LLMs
• Analyzes factual consistency without external resources
• Improves hallucination correction by 35% over baseline

Innovation

• Black-box sampling-based fact-level detection
• Knowledge graph representation of text facts
• Factual consistency analysis for hallucination scoring