Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

InFi-Check: Interpretable and Fine-Grained Fact-Checking of LLMs

Yuzhuo Bai, Shuzheng Si, Kangyang Luo, Qingyi Wang, Wenhao Li, Gang Chen, Fanchao Qi, Maosong Sun

TL;DR

InFi-Check is introduced, a framework for interpretable and fine-grained fact-checking of LLM outputs that achieves state-of-the-art performance on InFi-Check-FG and strong generalization across various downstream tasks, significantly improving the utility and trustworthiness of factuality evaluation.

Abstract

Large language models (LLMs) often hallucinate, yet most existing fact-checking methods treat factuality evaluation as a binary classification problem, offering limited interpretability and failing to capture fine-grained error types. In this paper, we introduce InFi-Check, a framework for interpretable and fine-grained fact-checking of LLM outputs. Specifically, we first propose a controlled data synthesis pipeline that generates high-quality data featuring explicit evidence, fine-grained error type labels, justifications, and corrections. Based on this, we further construct large-scale training data and a manually verified benchmark InFi-Check-FG for fine-grained fact-checking of LLM outputs. Building on these high-quality training data, we further propose InFi-Checker, which can jointly provide supporting evidence, classify fine-grained error types, and produce justifications along with corrections. Experiments show that InFi-Checker achieves state-of-the-art performance on InFi-Check-FG and strong generalization across various downstream tasks, significantly improving the utility and trustworthiness of factuality evaluation.

InFi-Check: Interpretable and Fine-Grained Fact-Checking of LLMs

TL;DR

InFi-Check is introduced, a framework for interpretable and fine-grained fact-checking of LLM outputs that achieves state-of-the-art performance on InFi-Check-FG and strong generalization across various downstream tasks, significantly improving the utility and trustworthiness of factuality evaluation.

Abstract

Large language models (LLMs) often hallucinate, yet most existing fact-checking methods treat factuality evaluation as a binary classification problem, offering limited interpretability and failing to capture fine-grained error types. In this paper, we introduce InFi-Check, a framework for interpretable and fine-grained fact-checking of LLM outputs. Specifically, we first propose a controlled data synthesis pipeline that generates high-quality data featuring explicit evidence, fine-grained error type labels, justifications, and corrections. Based on this, we further construct large-scale training data and a manually verified benchmark InFi-Check-FG for fine-grained fact-checking of LLM outputs. Building on these high-quality training data, we further propose InFi-Checker, which can jointly provide supporting evidence, classify fine-grained error types, and produce justifications along with corrections. Experiments show that InFi-Checker achieves state-of-the-art performance on InFi-Check-FG and strong generalization across various downstream tasks, significantly improving the utility and trustworthiness of factuality evaluation.
Paper Structure (32 sections, 10 figures, 13 tables)

This paper contains 32 sections, 10 figures, 13 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Controlled Data Synthesis Pipeline
  5. Fine-Grained Error Type Construction
  6. Semantic-Level Hallucinations.
  7. InFi-Check Dataset and Model
  8. Dataset and Benchmark
  9. Fact-Checking Model: InFi-Checker
  10. Experiment
  11. Experimental Setup
  12. InFi-Check-FG Results
  13. OOD Generalization Results
  14. In-depth Sentence Level Analysis
  15. Ablation Study
  16. ...and 17 more sections

Figures (10)

  • Figure 1: The illustration of our InFi-Check. InFi-Check can simultaneously provide the corresponding evidence, fine-grained labels, justifications, and corrections.
  • Figure 2: Overview of the InFi-Check pipeline. Some of the text is simplified for better demonstration.
  • Figure 3: Document and claim length (words) distribution of InFi-Check-FG with average length comparison.
  • Figure 4: An example of the data in InFi-Check-FG, the data is truncated due to space limitations.
  • Figure 5: Prompt for writing summaries.
  • ...and 5 more figures