Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GraphCheck: Multipath Fact-Checking with Entity-Relationship Graphs

Hyewon Jeon, Jay-Yoon Lee

TL;DR

GraphCheck reframes complex, multi-hop fact-checking as verification over entity-relationship graphs, enabling multipath reasoning by infilling latent entities and independently verifying each fact triplet. The authors augment this with DP-GraphCheck, a lightweight strategy selector that adaptively chooses between direct prompt-based verification and the graph-driven approach to balance accuracy and efficiency. Empirical results on HOVER and EX-FEVER show that GraphCheck and DP-GraphCheck outperform baselines in verification accuracy while maintaining favorable runtime and cost, with the strategy selector also boosting performance for other pipelines. The work demonstrates robustness across graph-construction models and establishes a general, extensible framework for scalable, open-book fact-checking in real-world settings.

Abstract

Automated fact-checking aims to assess the truthfulness of textual claims based on relevant evidence. However, verifying complex claims that require multi-hop reasoning remains a significant challenge. We propose GraphCheck, a novel framework that transforms claims into entity-relationship graphs for structured and systematic fact-checking. By explicitly modeling both explicit and latent entities and exploring multiple reasoning paths, GraphCheck enhances verification robustness. While GraphCheck excels in complex scenarios, it may be unnecessarily elaborate for simpler claims. To address this, we introduce DP-GraphCheck, a variant that employs a lightweight strategy selector to choose between direct prompting and GraphCheck adaptively. This selective mechanism improves both accuracy and efficiency by applying the appropriate level of reasoning to each claim. Experiments on the HOVER and EX-FEVER datasets demonstrate that our approach outperforms existing methods in verification accuracy, while achieving strong computational efficiency despite its multipath exploration. Moreover, the strategy selection mechanism in DP-GraphCheck generalizes well to other fact-checking pipelines, highlighting the broad applicability of our framework.

GraphCheck: Multipath Fact-Checking with Entity-Relationship Graphs

TL;DR

GraphCheck reframes complex, multi-hop fact-checking as verification over entity-relationship graphs, enabling multipath reasoning by infilling latent entities and independently verifying each fact triplet. The authors augment this with DP-GraphCheck, a lightweight strategy selector that adaptively chooses between direct prompt-based verification and the graph-driven approach to balance accuracy and efficiency. Empirical results on HOVER and EX-FEVER show that GraphCheck and DP-GraphCheck outperform baselines in verification accuracy while maintaining favorable runtime and cost, with the strategy selector also boosting performance for other pipelines. The work demonstrates robustness across graph-construction models and establishes a general, extensible framework for scalable, open-book fact-checking in real-world settings.

Abstract

Automated fact-checking aims to assess the truthfulness of textual claims based on relevant evidence. However, verifying complex claims that require multi-hop reasoning remains a significant challenge. We propose GraphCheck, a novel framework that transforms claims into entity-relationship graphs for structured and systematic fact-checking. By explicitly modeling both explicit and latent entities and exploring multiple reasoning paths, GraphCheck enhances verification robustness. While GraphCheck excels in complex scenarios, it may be unnecessarily elaborate for simpler claims. To address this, we introduce DP-GraphCheck, a variant that employs a lightweight strategy selector to choose between direct prompting and GraphCheck adaptively. This selective mechanism improves both accuracy and efficiency by applying the appropriate level of reasoning to each claim. Experiments on the HOVER and EX-FEVER datasets demonstrate that our approach outperforms existing methods in verification accuracy, while achieving strong computational efficiency despite its multipath exploration. Moreover, the strategy selection mechanism in DP-GraphCheck generalizes well to other fact-checking pipelines, highlighting the broad applicability of our framework.

Paper Structure

This paper contains 52 sections, 2 figures, 15 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Multi-hop Fact-Checking
  4. Latent Entity Identification
  5. Graph-based Fact-Checking
  6. Methodology
  7. GraphCheck: Graph Construction
  8. GraphCheck: Latent Entity Infilling
  9. Latent Entity Identification
  10. Multipath Exploration
  11. GraphCheck: Fact Triplet Verification
  12. Triplet Verification
  13. Path Verification
  14. Claim Verification
  15. DP-GraphCheck
  16. ...and 37 more sections

Figures (2)

  • Figure 1: Overview of GraphCheck. Compared to prior approaches, GraphCheck offers a comprehensive yet fine-grained claim decomposition. It also systematically explores multiple paths instead of following a fixed path. The overall process consists of three steps: (1) A claim is converted into a structured entity-relationship graph in which both explicit and latent entities are represented. (2) Latent entities are then identified through text infilling. Multiple infilling paths are explored via graph traversal, resulting in multiple infilled graphs. (3) For each infilled graph, every triplet is individually verified. The claim is predicted as Supported if at least one path yields a graph in which all triplets are verified as Supported; otherwise, the claim is predicted as Not Supported.
  • Figure 2: An example from the EX-FEVER dataset.