GISExplainer: On Explainability of Graph Neural Networks via Game-theoretic Interaction Subgraphs

TL;DR

GISExplainer presents a game-theoretic interaction framework to explain GNNs by uncovering connected explanatory subgraphs. It defines edge-level causal contributions and subgraph importance through Shapley-based coalitions, integrating positive and negative interactions, and builds explanations via a greedy, connectivity-constrained search. The method achieves superior Fidelity and Sparsity on synthetic and real datasets and demonstrates robustness to adversarial perturbations, offering practical utility for reliable GNN explanations. This approach advances explainability by explicitly modeling edge coalitions, linking causal attribution to connected subgraph structures with efficient coalition-sampling strategies.

Abstract

Explainability is crucial for the application of black-box Graph Neural Networks (GNNs) in critical fields such as healthcare, finance, cybersecurity, and more. Various feature attribution methods, especially the perturbation-based methods, have been proposed to indicate how much each node/edge contributes to the model predictions. However, these methods fail to generate connected explanatory subgraphs that consider the causal interaction between edges within different coalition scales, which will result in unfaithful explanations. In our study, we propose GISExplainer, a novel game-theoretic interaction based explanation method that uncovers what the underlying GNNs have learned for node classification by discovering human-interpretable causal explanatory subgraphs. First, GISExplainer defines a causal attribution mechanism that considers the game-theoretic interaction of multi-granularity coalitions in candidate explanatory subgraph to quantify the causal effect of an edge on the prediction. Second, GISExplainer assumes that the coalitions with negative effects on the predictions are also significant for model interpretation, and the contribution of the computation graph stems from the combined influence of both positive and negative interactions within the coalitions. Then, GISExplainer regards the explanation task as a sequential decision process, in which a salient edges is successively selected and connected to the previously selected subgraph based on its causal effect to form an explanatory subgraph, ultimately striving for better explanations. Additionally, an efficiency optimization scheme is proposed for the causal attribution mechanism through coalition sampling. Extensive experiments demonstrate that GISExplainer achieves better performance than state-of-the-art approaches w.r.t. two quantitative metrics: Fidelity and Sparsity.

Figures (7)

• Figure 1: Explainability of GNNs. (a) Top-K most relevant edges for explanatory subgraph construction. (b) Causal effect of an edge on model's predictions. (c) The interaction strength of candidate explanatory subgraph based on game-theoretic interaction of multi-granularity coalitions.
• Figure 2: Illustration of the proposed GISExplainer. (a) The process of generating an explanatory subgraph progressively. Here, the red node represents the target node $v_i$, and the nodes with blue dashed coil is the extension node. The yellow edges indicate the candidate edges that can be selected for updating the explanatory subgraph. (b) The decision-making process of causal screening for salient edges. The salient edge is selected for updating the explanatory subgraph based on the contributions of the candidate edges, which can be measured by the interaction strength of the corresponding explanatory subgraphs. $B(\cdot)$ indicates that the reward of the coalition, and the interaction strength $T(G_i)$ of a corresponding explanatory subgraph is the combined effects of the coalitions.
• Figure 3: Coalitions for game-theoretic interaction of subgraphs.
• Figure 4: Heuristic search for explanatory subgraph construction.
• Figure 5: Visualization of GISExplainer's explanatory subgraphs on synthetic datasets