Beyond Soft Masks: Hard-Perturbation Mixup Explainer for Robust GNN Explainability

2026-06-04

AI Summary

Existing graph neural network (GNN) explanation methods predominantly rely on soft masking, which struggles to fully eliminate label-irrelevant information, often introducing redundant structures and exacerbating out-of-distribution generalization issues, thereby compromising explanation fidelity. To address these limitations, this work proposes the HPME framework, which introduces, for the first time, a hard perturbation mechanism combined with a structure-level replacement-mixing strategy. By integrating the generalized graph information bottleneck principle with graph pooling techniques, HPME generates discrete and compact explanatory subgraphs. Extensive experiments on multiple synthetic and real-world datasets demonstrate that the proposed method significantly enhances both the fidelity and robustness of GNN explanations, effectively mitigates distributional shifts, and achieves state-of-the-art performance in GNN interpretability.

Abstract

Graph Neural Networks (GNNs) have demonstrated remarkable performance across a range of applications involving graph-structured data, particularly in high-stakes domains. However, the opaque nature of their decision-making processes limits their trustworthiness and broader adoption. Existing post-hoc explanation methods aim to improve explainability by identifying subgraphs that influence GNN predictions and adopt mixup strategies to alleviate the out-of-distribution (OOD) issue caused by using subgraphs for prediction. Yet, these approaches typically rely on soft masks, which are inherently unable to fully eliminate label-irrelevant information, allowing redundant structures to leak into the mixup process and hindering the resolution of the OOD problem, thereby degrading explanation fidelity. In this work, we propose HPME, a Hard-Perturbation Mixup Explanation framework grounded in a generalized Graph Information Bottleneck, which leverages graph pooling to extract discrete explanatory subgraphs and to yield an information-capacity bound to thoroughly compress label-irrelevant components. Furthermore, we introduce a novel mixup strategy built upon structure-level replacement, generating in-distribution explanations to effectively mitigate the distribution shift. Extensive experiments on diverse tasks demonstrate that HPME achieves state-of-the-art performance in generating robust and interpretable explanations across both synthetic and real-world datasets.

Problem

• GNN explainability
• out-of-distribution
• soft masks
• explanation fidelity
• subgraph explanation

Innovation

• Hard-Perturbation Mixup
• Graph Information Bottleneck
• Graph Pooling
• In-Distribution Explanation
• GNN Explainability