Why Self-Inconsistency Arises in GNN Explanations and How to Exploit It

TL;DR

This paper investigates the causes of self-inconsistency in GNN explanations, introduces a hypothesis and a post-processing method called Self-Denoising to improve explanation stability and quality.

Contribution

It identifies re-explanation-induced context perturbation as the cause of self-inconsistency and proposes a simple, model-agnostic post-processing method to enhance explanation reliability.

Findings

Self-Denoising improves explanation quality across various GNN frameworks.

Self-Denoising adds only 4-6% computational overhead.

Re-explanation-induced context perturbation causes explanation score variation.

Abstract

Recent work has observed that explanations produced by Self-Interpretable Graph Neural Networks (SI-GNNs) can be self-inconsistent: when the model is reapplied to its own explanatory graph subset, it may produce a different explanation. However, why self-inconsistency arises remains poorly understood. In this work, we first identify re-explanation-induced context perturbation as the direct cause of score variation. We then introduce a latent signal assignment hypothesis to explain why only some edges are sensitive to this perturbation, and analyze how conciseness regularization affects latent signal assignment. Given that self-inconsistent edges do not provide stable evidence for the model's prediction, we propose Self-Denoising (SD), a model-agnostic and training-free post-processing strategy that calibrates explanations with only one additional forward pass. Experiments across representative SI-GNN frameworks, backbone architectures, and benchmark datasets support our hypothesis and show that SD consistently improves explanation quality while adding only about 4--6\% computational overhead in practice.