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

2026-05-08

AI Summary

This work addresses the unexplained inconsistency observed in self-explaining graph neural networks (SI-GNNs) when reinterpreting their generated subgraphs. The study reveals that this self-inconsistency stems from context perturbations induced by the re-interpretation process and introduces the “latent signal allocation” hypothesis to account for varying edge sensitivities. Building on this insight, the authors propose Self-Denoising—a training-free, model-agnostic post-processing method that calibrates explanations with just one additional forward pass. Theoretical analysis and extensive experiments demonstrate that Self-Denoising substantially enhances explanation stability and quality across diverse SI-GNN architectures, backbone networks, and benchmark datasets, while incurring only a marginal computational overhead of 4–6%.

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.

Problem

• self-inconsistency
• GNN explanations
• context perturbation
• latent signal assignment
• explanation stability

Innovation

• self-inconsistency
• graph neural networks
• explanation calibration
• context perturbation
• Self-Denoising