Overcoming Shortcut Learning in Graph Neural Networks through Active Explanation Guidance

2026-08-14

AI Summary

This study addresses the reliability degradation of Graph Neural Networks in out-of-distribution tasks caused by reliance on non-causal shortcuts. We propose XIGL, an architecture-agnostic human-in-the-loop strategy that innovatively integrates active learning with explainability analysis. By optimizing query strategies to prioritize the identification and correction of shortcut behaviors within explanations, XIGL efficiently resolves model confusion through expert feedback. Extensive experiments demonstrate that this approach effectively suppresses shortcut learning across diverse GNN architectures, significantly enhancing both out-of-distribution generalization and prediction reliability. Consequently, this work establishes a novel paradigm for constructing robust graph learning systems capable of mitigating spurious correlations through interpretable human-AI collaboration.

Abstract

Graph Neural Networks (GNNs) can solve prediction tasks by unintentionally exploiting shortcuts---that is, edges, nodes, and features that correlate with but are not causal for the prediction---which compromise their reliability in out-of-distribution tasks. We introduce XIGL, an architecture-agnostic human-in-the-loop strategy for removing such shortcuts from GNNs. Our key insight is twofold. On the one hand, reliance on shortcuts can be detected by inspecting GNN explanations. On the other hand, once made aware of such shortcuts, sufficiently expert users can provide tailored corrective feedback, which helps deconfound the model. XIGL supports any query strategy; however, since corrective feedback can be expensive to acquire, we develop an active learning strategy for prioritizing explanations that are more likely to display shortcut behavior, lowering annotation and cognitive costs. We showcase the effectiveness of XIGL, including both existing and proposed explanation-based strategies, on several GNN architectures. Our implementation is available online.

Problem

• Shortcut Learning
• Graph Neural Networks
• Out-of-Distribution
• Reliability

Innovation

• Shortcut Learning
• Graph Neural Networks
• Human-in-the-loop
• Active Learning
• Explainability