GnnXemplar: Exemplars to Explanations - Natural Language Rules for Global GNN Interpretability

2025-09-22

AI Summary

Graph Neural Networks (GNNs) suffer from limited global interpretability in node classification, especially on large-scale, high-dimensional graphs with low subgraph redundancy—where existing subgraph-pattern-based explanation methods fail. To address this, we propose GnnXemplar: the first framework to leverage reverse k-nearest-neighbor coverage maximization for selecting representative nodes in the GNN embedding space, integrated with a large language model (LLM)-driven self-optimizing prompting strategy to generate human-readable natural-language rules that expose GNN decision logic. This end-to-end approach jointly models structure-attribute interactions while ensuring human comprehensibility. Extensive experiments on multiple benchmark datasets demonstrate superior explanation fidelity, scalability, and readability over state-of-the-art baselines. A user study with 60 participants further confirms the effectiveness and practical utility of GnnXemplar’s explanations.

Technology Category

• Humans and AI: Explainable AI (XAI) for Human Understanding
• Machine Learning: Graph-based Machine Learning
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

Graph Neural Networks (GNNs) are widely used for node classification, yet their opaque decision-making limits trust and adoption. While local explanations offer insights into individual predictions, global explanation methods, those that characterize an entire class, remain underdeveloped. Existing global explainers rely on motif discovery in small graphs, an approach that breaks down in large, real-world settings where subgraph repetition is rare, node attributes are high-dimensional, and predictions arise from complex structure-attribute interactions. We propose GnnXemplar, a novel global explainer inspired from Exemplar Theory from cognitive science. GnnXemplar identifies representative nodes in the GNN embedding space, exemplars, and explains predictions using natural language rules derived from their neighborhoods. Exemplar selection is framed as a coverage maximization problem over reverse k-nearest neighbors, for which we provide an efficient greedy approximation. To derive interpretable rules, we employ a self-refining prompt strategy using large language models (LLMs). Experiments across diverse benchmarks show that GnnXemplar significantly outperforms existing methods in fidelity, scalability, and human interpretability, as validated by a user study with 60 participants.

Problem

• Explaining opaque decision-making in Graph Neural Networks for node classification
• Addressing limitations of global explanation methods for entire class characterization
• Overcoming motif discovery breakdown in large real-world graphs with complex interactions

Innovation

• Exemplar selection via reverse k-nearest neighbors coverage
• Natural language rules generated using self-refining LLM prompts
• Global GNN explanations from representative nodes' neighborhoods