KEA Explain: Explanations of Hallucinations using Graph Kernel Analysis

2025-07-04

AI Summary

Large language models (LLMs) frequently generate factually incorrect outputs—termed “hallucinations”—posing critical risks in high-stakes domains. Method: This paper proposes a neuro-symbolic framework that first structures LLM outputs into knowledge graphs, then applies graph kernel computation and semantic clustering to compare them against external knowledge sources (e.g., Wikidata or context documents), enabling automated hallucination detection and interpretable attribution. Contribution/Results: To our knowledge, this is the first work integrating graph kernels with semantic clustering for hallucination explanation. The approach achieves state-of-the-art detection accuracy in both open-domain and closed-domain settings, while delivering fine-grained, contrastive attributions. It exhibits strong cross-domain generalization, high interpretability, and robust factual consistency. Empirical evaluation demonstrates significant improvements in trustworthiness and transparency of LLMs in safety-critical applications such as healthcare and legal reasoning.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• 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
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Large Language Models (LLMs) frequently generate hallucinations: statements that are syntactically plausible but lack factual grounding. This research presents KEA (Kernel-Enriched AI) Explain: a neurosymbolic framework that detects and explains such hallucinations by comparing knowledge graphs constructed from LLM outputs with ground truth data from Wikidata or contextual documents. Using graph kernels and semantic clustering, the method provides explanations for detected hallucinations, ensuring both robustness and interpretability. Our framework achieves competitive accuracy in detecting hallucinations across both open- and closed-domain tasks, and is able to generate contrastive explanations, enhancing transparency. This research advances the reliability of LLMs in high-stakes domains and provides a foundation for future work on precision improvements and multi-source knowledge integration.

Problem

• Detect and explain hallucinations in LLM outputs
• Compare knowledge graphs with ground truth data
• Enhance LLM reliability in high-stakes domains

Innovation

• Neurosymbolic framework detects LLM hallucinations
• Graph kernel analysis compares knowledge graphs
• Semantic clustering ensures robust interpretability