Concept Graph Convolutions: Message Passing in the Concept Space

TL;DR

This paper introduces Concept Graph Convolutions, a novel graph convolution method operating on node-level concepts to enhance interpretability and provide insights into message passing processes.

Contribution

It presents the first graph convolution designed for node-level concepts, improving interpretability and maintaining competitive accuracy.

Findings

Enables understanding of concept evolution across layers

Achieves competitive accuracy on graph tasks

Provides a pure concept space convolution variant

Abstract

The trust in the predictions of Graph Neural Networks is limited by their opaque reasoning process. Prior methods have tried to explain graph networks via concept-based explanations extracted from the latent representations obtained after message passing. However, these explanations fall short of explaining the message passing process itself. To this aim, we propose the Concept Graph Convolution, the first graph convolution designed to operate on node-level concepts for improved interpretability. The proposed convolutional layer performs message passing on a combination of raw and concept representations using structural and attention-based edge weights. We also propose a pure variant of the convolution, only operating in the concept space. Our results show that the Concept Graph Convolution allows to obtain competitive task accuracy, while enabling an increased insight into the evolution of concepts across convolutional steps.