GInX-Eval: Towards In-Distribution Evaluation of Graph Neural Network Explanations

TL;DR

GInX-Eval introduces a new in-distribution evaluation framework for graph neural network explanations, addressing out-of-distribution issues and providing more reliable metrics that align with human judgment.

Contribution

The paper proposes GInX-Eval, a novel evaluation method for GNN explanations that overcomes faithfulness limitations and assesses explanation quality in the model's distribution.

Findings

Many popular explanation methods are no better than random.

GInX-Eval metrics align with human evaluation.

The approach is consistent across multiple datasets.

Abstract

Diverse explainability methods of graph neural networks (GNN) have recently been developed to highlight the edges and nodes in the graph that contribute the most to the model predictions. However, it is not clear yet how to evaluate the correctness of those explanations, whether it is from a human or a model perspective. One unaddressed bottleneck in the current evaluation procedure is the problem of out-of-distribution explanations, whose distribution differs from those of the training data. This important issue affects existing evaluation metrics such as the popular faithfulness or fidelity score. In this paper, we show the limitations of faithfulness metrics. We propose GInX-Eval (Graph In-distribution eXplanation Evaluation), an evaluation procedure of graph explanations that overcomes the pitfalls of faithfulness and offers new insights on explainability methods. Using a fine-tuning strategy, the GInX score measures how informative removed edges are for the model and the EdgeRank score evaluates if explanatory edges are correctly ordered by their importance. GInX-Eval verifies if ground-truth explanations are instructive to the GNN model. In addition, it shows that many popular methods, including gradient-based methods, produce explanations that are not better than a random designation of edges as important subgraphs, challenging the findings of current works in the area. Results with GInX-Eval are consistent across multiple datasets and align with human evaluation.