SliceGX: Layer-wise GNN Explanation with Model-slicing

2025-06-22

AI Summary

Existing GNN explanation methods predominantly rely on input perturbation to identify output-influential subgraphs, failing to characterize the layer-wise propagation of intermediate representations—thus hindering model diagnosis and architectural optimization. To address this, we propose SliceGX, the first GNN interpretability framework supporting progressive, layer-wise explanation. SliceGX decouples GNNs layer-by-layer via model slicing and dynamically extracts high-fidelity explanatory subgraphs within each layer block using an incremental subgraph generation and maintenance algorithm. It further introduces a SPARQL-like declarative query interface for flexible retrieval and analysis, backed by theoretical approximation guarantees. Extensive experiments on large-scale real-world graphs and mainstream GNN architectures demonstrate that SliceGX significantly improves both explanation accuracy and efficiency. It enables fine-grained, verifiable, layer-specific explanations—facilitating GNN debugging, attribution analysis, and structural optimization.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Humans and AI: Explainable AI (XAI) for Human Understanding

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

Ensuring the trustworthiness of graph neural networks (GNNs) as black-box models requires effective explanation methods. Existing GNN explanations typically apply input perturbations to identify subgraphs that are responsible for the occurrence of the final output of GNNs. However, such approaches lack finer-grained, layer-wise analysis of how intermediate representations contribute to the final result, capabilities that are crucial for model diagnosis and architecture optimization. This paper introduces SliceGX, a novel GNN explanation approach that generates explanations at specific GNN layers in a progressive manner. Given a GNN M, a set of selected intermediate layers, and a target layer, SliceGX automatically segments M into layer blocks ("model slice") and discovers high-quality explanatory subgraphs in each layer block that clarifies the occurrence of output of M at the targeted layer. Although finding such layer-wise explanations is computationally challenging, we develop efficient algorithms and optimization techniques that incrementally generate and maintain these subgraphs with provable approximation guarantees. Additionally, SliceGX offers a SPARQL-like query interface, providing declarative access and search capacities for the generated explanations. Through experiments on large real-world graphs and representative GNN architectures, we verify the effectiveness and efficiency of SliceGX, and illustrate its practical utility in supporting model debugging.

Problem

• Lack of layer-wise GNN explanation methods
• Need for intermediate representation analysis in GNNs
• Computational challenges in generating layer-wise explanations

Innovation

• Layer-wise GNN explanation with model-slicing
• Efficient algorithms with approximation guarantees
• SPARQL-like query interface for explanations