PGM-Explainer: Probabilistic Graphical Model Explanations for Graph Neural Networks

In Graph Neural Networks (GNNs), the graph structure is incorporated into the of node representations. This complex structure makes explaining GNNs' become much more challenging. In this paper, we propose-Explainer, a Probabilistic Graphical Model (PGM) model-agnostic explainer GNNs. Given a prediction to be explained, PGM-Explainer identifies crucial components and generates an explanation in form of a PGM approximating prediction. Different from existing explainers for GNNs where the are drawn from a set of linear functions of explained features, -Explainer is able to demonstrate the dependencies of explained features in of conditional probabilities. Our theoretical analysis shows that the PGM by PGM-Explainer includes the Markov-blanket of the target, i. e. including all its statistical information. We also show that explanation returned by PGM-Explainer contains the same set of independence in the perfect map. Our experiments on both synthetic and real-world show that PGM-Explainer achieves better performance than existing in many benchmark tasks.