MuPlon: Multi-Path Causal Optimization for Claim Verification through Controlling Confounding

TL;DR

MuPlon addresses claim verification under data noise and biases by integrating dual causal interventions on a fully connected Claim-Evidence Graph. The back-door path adjustment dilutes noisy evidence via a Bayesian node-sampling approach and enhances local evidence cohesion, while the front-door path adjustment constructs mediator-based reasoning paths and applies counterfactual reasoning to remove biases. Empirical results across FEVER, Politihop, and Cladder show state-of-the-art accuracy and robustness, including adversarial and symmetric settings, demonstrating the practical value of principled causal interventions in evidence-based verification. The approach offers a scalable, principled framework for reliable verification in noisy, bias-prone data environments, with potential for broader domain adaptation and real-time reasoning.

Abstract

As a critical task in data quality control, claim verification aims to curb the spread of misinformation by assessing the truthfulness of claims based on a wide range of evidence. However, traditional methods often overlook the complex interactions between evidence, leading to unreliable verification results. A straightforward solution represents the claim and evidence as a fully connected graph, which we define as the Claim-Evidence Graph (C-E Graph). Nevertheless, claim verification methods based on fully connected graphs face two primary confounding challenges, Data Noise and Data Biases. To address these challenges, we propose a novel framework, Multi-Path Causal Optimization (MuPlon). MuPlon integrates a dual causal intervention strategy, consisting of the back-door path and front-door path. In the back-door path, MuPlon dilutes noisy node interference by optimizing node probability weights, while simultaneously strengthening the connections between relevant evidence nodes. In the front-door path, MuPlon extracts highly relevant subgraphs and constructs reasoning paths, further applying counterfactual reasoning to eliminate data biases within these paths. The experimental results demonstrate that MuPlon outperforms existing methods and achieves state-of-the-art performance.

Figures (3)

• Figure 1: Construction of the C-E Graph and the issues of data biases and noise. Red dashed lines indicate irrelevant edges.
• Figure 2:
• Figure 3: An overview of MuPlon Framework (The gray part is the causal intervention theory framework).