Link Prediction with Physics-Inspired Graph Neural Networks

2024-02-22

AI Summary

To address the challenges of large label disparities among adjacent nodes in heterogeneous graphs and the unsuitability of conventional GNN message-passing mechanisms for link prediction, this paper proposes GRAFF-LP, a physics-inspired model. Methodologically, it introduces a physically grounded bias mechanism—first systematically applied to link prediction—and designs an edge-gradient-decoupled, physics-driven readout function, revealing the need for specialized evaluation metrics in heterogeneous link prediction. Furthermore, it enhances heterogeneous graph representation via implicit edge-gradient learning. Experiments demonstrate that GRAFF-LP significantly improves link prediction performance across diverse GNN backbones on heterogeneous graphs. Notably, simple GNNs exhibit comparable prediction difficulty on heterogeneous versus homogeneous graphs, confirming that modeling bias—not architectural complexity—is the fundamental bottleneck. The code is publicly available to ensure full reproducibility.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Reasoning under Uncertainty: Graphical Models
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs 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

The message-passing mechanism underlying Graph Neural Networks (GNNs) is not naturally suited for heterophilic datasets, where adjacent nodes often have different labels. Most solutions to this problem remain confined to the task of node classification. In this article, we focus on the valuable task of link prediction under heterophily, an interesting problem for recommendation systems, social network analysis, and other applications. GNNs like GRAFF have improved node classification under heterophily by incorporating physics biases in the architecture. Similarly, we propose GRAFF-LP, an extension of GRAFF for link prediction. We show that GRAFF-LP effectively discriminates existing from non-existing edges by learning implicitly to separate the edge gradients. Based on this information, we propose a new readout function inspired by physics. Remarkably, this new function not only enhances the performance of GRAFF-LP but also improves that of other baseline models, leading us to reconsider how every link prediction experiment has been conducted so far. Finally, we provide evidence that even simple GNNs did not experience greater difficulty in predicting heterophilic links compared to homophilic ones. This leads us to believe in the necessity for heterophily measures specifically tailored for link prediction, distinct from those used in node classification. The code for reproducing our experiments is available at this URL https://anonymous.4open.science/r/Link_Prediction_with_PIGNN_IJCNN-F03F/.

Problem

• Improving link prediction in heterophilic graphs using GNNs
• Proposing GRAFF-LP for edge discrimination via gradient separation
• Introducing physics-inspired readout to enhance model performance

Innovation

• Extends GRAFF for link prediction under heterophily
• Learns to separate edge gradients implicitly
• Introduces physics-inspired readout function