Interpreting Graph Neural Networks for NLP With Differentiable Edge Masking

TL;DR

This paper presents a differentiable, post-hoc interpretability method for GNNs in NLP that identifies and drops unnecessary edges, revealing insights into model decision processes without sacrificing performance.

Contribution

Introduces a novel differentiable edge masking approach for interpreting GNNs in NLP, enabling analysis of information flow and importance of graph components.

Findings

Large proportion of edges can be dropped without performance loss

The method provides meaningful insights into model decision-making

Edges identified as important align with linguistic structures

Abstract

Graph neural networks (GNNs) have become a popular approach to integrating structural inductive biases into NLP models. However, there has been little work on interpreting them, and specifically on understanding which parts of the graphs (e.g. syntactic trees or co-reference structures) contribute to a prediction. In this work, we introduce a post-hoc method for interpreting the predictions of GNNs which identifies unnecessary edges. Given a trained GNN model, we learn a simple classifier that, for every edge in every layer, predicts if that edge can be dropped. We demonstrate that such a classifier can be trained in a fully differentiable fashion, employing stochastic gates and encouraging sparsity through the expected $L_0$ norm. We use our technique as an attribution method to analyze GNN models for two tasks -- question answering and semantic role labeling -- providing insights into the information flow in these models. We show that we can drop a large proportion of edges without deteriorating the performance of the model, while we can analyse the remaining edges for interpreting model predictions.