Accumulated Local Effects and Graph Neural Networks for link prediction

2025-11-25

AI Summary

Existing Accumulated Local Effects (ALE) methods cannot be directly applied to link prediction in Graph Neural Networks (GNNs) due to nonlinear, high-order feature interactions among nodes induced by message passing. Method: We propose a scalable approximate ALE computation framework for GNNs, leveraging local subgraph sampling and gradient proxy strategies to preserve the causal semantics of ALE while drastically reducing computational complexity. Contribution/Results: Experiments on GCN and GAT demonstrate that the approximate ALE achieves explanation fidelity statistically indistinguishable from exact ALE (p > 0.05), with inference speedups of 1–2 orders of magnitude. Exact ALE exhibits greater robustness under limited samples, revealing an inherent trade-off between interpretability and efficiency. To our knowledge, this is the first work to systematically adapt ALE to GNN-based link prediction, providing an efficient, reliable, and model-agnostic visualization tool for local attribution in graph models.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Reasoning under Uncertainty: Graphical Models
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

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 learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

We investigate how Accumulated Local Effects (ALE), a model-agnostic explanation method, can be adapted to visualize the influence of node feature values in link prediction tasks using Graph Neural Networks (GNNs), specifically Graph Convolutional Networks and Graph Attention Networks. A key challenge addressed in this work is the complex interactions of nodes during message passing within GNN layers, complicating the direct application of ALE. Since a straightforward solution of modifying only one node at once substantially increases computation time, we propose an approximate method that mitigates this challenge. Our findings reveal that although the approximate method offers computational efficiency, the exact method yields more stable explanations, particularly when smaller data subsets are used. However, the explanations produced with the approximate method are not significantly different from the ones obtained with the exact method. Additionally, we analyze how varying parameters affect the accuracy of ALE estimation for both approaches.

Problem

• Adapting ALE to visualize node feature influence in GNN link prediction
• Addressing computational challenges in applying ALE to GNN message passing
• Comparing exact and approximate ALE methods for explanation stability and accuracy

Innovation

• Adapt ALE for GNN link prediction visualization
• Propose approximate method to reduce computation time
• Compare exact and approximate ALE stability and accuracy