FactCHD: Benchmarking Fact-Conflicting Hallucination Detection

TL;DR

FactCHD is a new benchmark for detecting fact-conflicting hallucinations in LLM outputs, highlighting current detection challenges and proposing a tool-enhanced method for improved factuality assessment.

Contribution

We introduce FactCHD, a comprehensive benchmark with evidence chains for evaluating hallucination detection, and propose Truth-Triangulator, a novel approach combining multiple tools for better factuality detection.

Findings

Current methods struggle with accurate hallucination detection.

FactCHD reveals significant gaps in existing detection approaches.

Truth-Triangulator improves detection credibility by integrating evidence and predictions.

Abstract

Despite their impressive generative capabilities, LLMs are hindered by fact-conflicting hallucinations in real-world applications. The accurate identification of hallucinations in texts generated by LLMs, especially in complex inferential scenarios, is a relatively unexplored area. To address this gap, we present FactCHD, a dedicated benchmark designed for the detection of fact-conflicting hallucinations from LLMs. FactCHD features a diverse dataset that spans various factuality patterns, including vanilla, multi-hop, comparison, and set operation. A distinctive element of FactCHD is its integration of fact-based evidence chains, significantly enhancing the depth of evaluating the detectors' explanations. Experiments on different LLMs expose the shortcomings of current approaches in detecting factual errors accurately. Furthermore, we introduce Truth-Triangulator that synthesizes reflective considerations by tool-enhanced ChatGPT and LoRA-tuning based on Llama2, aiming to yield more credible detection through the amalgamation of predictive results and evidence. The benchmark dataset is available at https://github.com/zjunlp/FactCHD.