Discovering Invariant Rationales for Graph Neural Networks

TL;DR

This paper introduces a method called DIR that enhances the interpretability and robustness of graph neural networks by discovering invariant rationales across different data distributions, reducing reliance on spurious features.

Contribution

The paper proposes a novel intervention-based approach to identify causal, invariant rationales in GNNs, improving interpretability and out-of-distribution generalization.

Findings

DIR outperforms baselines in interpretability

DIR improves generalization on OOD data

Experiments on synthetic and real datasets validate effectiveness

Abstract

Intrinsic interpretability of graph neural networks (GNNs) is to find a small subset of the input graph's features -- rationale -- which guides the model prediction. Unfortunately, the leading rationalization models often rely on data biases, especially shortcut features, to compose rationales and make predictions without probing the critical and causal patterns. Moreover, such data biases easily change outside the training distribution. As a result, these models suffer from a huge drop in interpretability and predictive performance on out-of-distribution data. In this work, we propose a new strategy of discovering invariant rationale (DIR) to construct intrinsically interpretable GNNs. It conducts interventions on the training distribution to create multiple interventional distributions. Then it approaches the causal rationales that are invariant across different distributions while filtering out the spurious patterns that are unstable. Experiments on both synthetic and real-world datasets validate the superiority of our DIR in terms of interpretability and generalization ability on graph classification over the leading baselines. Code and datasets are available at https://github.com/Wuyxin/DIR-GNN.