LRP4RAG: Detecting Hallucinations in Retrieval-Augmented Generation via Layer-wise Relevance Propagation

TL;DR

LRP4RAG introduces a novel Layer-wise Relevance Propagation-based approach to detect hallucinations in Retrieval-Augmented Generation, significantly improving accuracy over existing methods.

Contribution

This paper pioneers the use of LRP for hallucination detection in RAG, combining relevance analysis with classifiers for better identification.

Findings

LRP4RAG outperforms existing baselines in hallucination detection

Relevance-based features improve detection accuracy

First application of LRP in RAG hallucination detection

Abstract

Retrieval-Augmented Generation (RAG) has become a primary technique for mitigating hallucinations in large language models (LLMs). However, incomplete knowledge extraction and insufficient understanding can still mislead LLMs to produce irrelevant or even contradictory responses, which means hallucinations persist in RAG. In this paper, we propose LRP4RAG, a method based on the Layer-wise Relevance Propagation (LRP) algorithm for detecting hallucinations in RAG. Specifically, we first utilize LRP to compute the relevance between the input and output of the RAG generator. We then apply further extraction and resampling to the relevance matrix. The processed relevance data are input into multiple classifiers to determine whether the output contains hallucinations. To the best of our knowledge, this is the first time that LRP has been used for detecting RAG hallucinations, and extensive experiments demonstrate that LRP4RAG outperforms existing baselines.