Multi-Path Causal Optimization for claim verification through controlling confounding

Hanghui Guo

doi:10.1016/j.engappai.2026.116117

Multi-Path Causal Optimization for claim verification through controlling confounding

2026-09-01

0

PRE

AI

DOI：10.1016/j.engappai.2026.116117

Original

Original request for help

Save

Abstract

En 中文

As an important task in information quality control systems, claim verification aims to curb misinformation by assessing the truthfulness of claims based on multiple pieces of evidence. However, traditional methods often overlook the complex interactions among evidence sources, leading to unreliable verification results. A straightforward solution is to represent claims and evidence as a fully connected graph. Nevertheless, graph-based claim verification methods often face two primary confounding challenges: data noise and data bias. To address these issues, we propose a novel method named Multi-Path Causal Optimization (MuPlon). MuPlon integrates dual causal intervention strategies using both the back-door path and the front-door path. In the back-door path, MuPlon suppresses noisy node interference by optimizing node probability weights while strengthening the connections among relevant evidence nodes. In the front-door path, MuPlon extracts highly relevant subgraphs, constructs reasoning paths, and applies counterfactual reasoning to mitigate data bias within these paths. Experimental results show that MuPlon achieves competitive and robust performance compared with existing claim verification methods while remaining practical under local deployment settings. These findings highlight the potential of causal graph reasoning for real-world data quality control systems and engineering information verification applications.

Multi-Path Causal Optimization for claim verification through controlling confounding

Abstract

Journal

Organization