Enhancing graph neural networks for self-explainable modeling: A causal perspective with multi-granularity receptive fields

Yuan Li, Li Liu, Penggang Chen, Chenglin Zhang, Guoyin Wang

Abstract: Highlights•Propose a multi-granularity receptive field to help the acquisition of causal correlations.•Design sliced architecture to integrate graph embedding with adaptive weights for explanations.•Design loss functions to get causality to ensure performance and reduce non-causal part impact.•Conduct on five real-world and one synthetic dataset in graph classification and explanation.