Technology Category
Machine Learning: Graph-based Machine LearningReasoning under Uncertainty: Graphical ModelsComputer Vision: Interpretability, Explainability, and Transparency
Ignoring Directionality Leads to Compromised Graph Neural Network Explanations
🤖 AI Summary
Graph Neural Networks (GNNs) require high-fidelity explanations in safety-critical domains, yet conventional graph symmetrization—commonly applied to handle undirected graphs—ignores edge directionality, inducing explanation distortion and degrading fidelity. Method: This work is the first to systematically expose the severe adverse impact of directionality loss on GNN interpretability; it introduces a direction-aware explanation evaluation framework with theoretical guarantees and proposes a directed-graph–specific sensitivity analysis and attribution method—namely, an enhanced GNNExplainer. Results: Extensive evaluation across multiple benchmark datasets demonstrates that preserving edge directionality improves explanation faithfulness by up to 42%, substantially enhancing decision trustworthiness in safety-sensitive applications. The core contribution is establishing directionality preservation as a necessary condition for faithful GNN interpretability and delivering the first theoretically rigorous and empirically validated direction-aware evaluation paradigm.
Technology Category
Application Category
📝 Abstract
Graph Neural Networks (GNNs) are increasingly used in critical domains, where reliable explanations are vital for supporting human decision-making. However, the common practice of graph symmetrization discards directional information, leading to significant information loss and misleading explanations. Our analysis demonstrates how this practice compromises explanation fidelity. Through theoretical and empirical studies, we show that preserving directional semantics significantly improves explanation quality, ensuring more faithful insights for human decision-makers. These findings highlight the need for direction-aware GNN explainability in security-critical applications.
Problem
Research questions and friction points this paper is trying to address.
Graph symmetrization discards directional information in GNNs
Directional information loss leads to misleading GNN explanations
Preserving directional semantics improves explanation fidelity
Innovation
Methods, ideas, or system contributions that make the work stand out.
Preserves directional semantics in GNNs
Improves explanation quality significantly
Ensures faithful insights for decision-makers
🔎 Similar Papers
No similar papers found.
