Explaining Graph Neural Networks for Node Similarity on Graphs

TL;DR

This paper investigates explainable methods for GNN-based node similarity search, comparing mutual information and gradient-based explanations, and finds gradient-based explanations to have desirable properties such as actionability and sparsity.

Contribution

It evaluates and compares two explanation methods for GNN-based similarity search, highlighting the advantages of gradient-based explanations over mutual information explanations.

Findings

Gradient-based explanations are actionable, leading to predictable similarity score changes.

Gradient explanations are consistent, with minimal overlap in input effects.

Gradient explanations can be pruned to produce sparse, effective explanations.

Abstract

Similarity search is a fundamental task for exploiting information in various applications dealing with graph data, such as citation networks or knowledge graphs. While this task has been intensively approached from heuristics to graph embeddings and graph neural networks (GNNs), providing explanations for similarity has received less attention. In this work we are concerned with explainable similarity search over graphs, by investigating how GNN-based methods for computing node similarities can be augmented with explanations. Specifically, we evaluate the performance of two prominent approaches towards explanations in GNNs, based on the concepts of mutual information (MI), and gradient-based explanations (GB). We discuss their suitability and empirically validate the properties of their explanations over different popular graph benchmarks. We find that unlike MI explanations, gradient-based explanations have three desirable properties. First, they are actionable: selecting inputs depending on them results in predictable changes in similarity scores. Second, they are consistent: the effect of selecting certain inputs overlaps very little with the effect of discarding them. Third, they can be pruned significantly to obtain sparse explanations that retain the effect on similarity scores.