Global Concept-Based Interpretability for Graph Neural Networks via Neuron Analysis

TL;DR

This paper introduces a novel global interpretability method for GNNs by analyzing individual neuron behaviors, revealing that neurons act as concept detectors aligned with logical graph properties, and offers a more transparent explanation framework.

Contribution

It is the first to demonstrate that GNN neurons function as concept detectors and to develop a global explanation approach based on neuron-level concepts.

Findings

GNN neurons act as concept detectors aligned with logical graph properties

A trade-off exists between training duration and neuron interpretability

The proposed global explanation method improves interpretability and reduces bias

Abstract

Graph neural networks (GNNs) are highly effective on a variety of graph-related tasks; however, they lack interpretability and transparency. Current explainability approaches are typically local and treat GNNs as black-boxes. They do not look inside the model, inhibiting human trust in the model and explanations. Motivated by the ability of neurons to detect high-level semantic concepts in vision models, we perform a novel analysis on the behaviour of individual GNN neurons to answer questions about GNN interpretability, and propose new metrics for evaluating the interpretability of GNN neurons. We propose a novel approach for producing global explanations for GNNs using neuron-level concepts to enable practitioners to have a high-level view of the model. Specifically, (i) to the best of our knowledge, this is the first work which shows that GNN neurons act as concept detectors and have strong alignment with concepts formulated as logical compositions of node degree and neighbourhood properties; (ii) we quantitatively assess the importance of detected concepts, and identify a trade-off between training duration and neuron-level interpretability; (iii) we demonstrate that our global explainability approach has advantages over the current state-of-the-art -- we can disentangle the explanation into individual interpretable concepts backed by logical descriptions, which reduces potential for bias and improves user-friendliness.