Boundary Embedding Shaping with Adaptive Contrastive Learning for Graph Structural Disentanglement

2026-06-18

AI Summary

Graph neural networks are prone to interference from irrelevant neighbors in decision boundary regions, leading to distorted embeddings of boundary nodes and degraded classification performance. This work addresses this structural entanglement issue—previously unexplored—and proposes Boundary Embedding Shaping (BES), a novel approach that leverages adaptive contrastive learning and a boundary-aware mechanism to selectively suppress spurious structural noise. Implemented as a lightweight plug-in module, BES achieves significant embedding refinement with minimal parameter perturbation. Experimental results demonstrate that BES substantially enhances model discriminability, yielding an average 3.3% absolute improvement over GCN on node classification tasks—with gains up to 5.0% on WikiCS—and also achieves superior accuracy in link prediction.

Abstract

Graph neural networks (GNNs) excel at aggregating neighbor information for classification, yet their performance is hindered by graph structural entanglement, where spurious correlations from semantically irrelevant neighbors contaminate node embeddings. This challenge is most acute for nodes near class boundaries in the embedding space, where amplified structural noise blurs decision boundaries and destabilizes predictions. Existing robust GNN methods largely treat all nodes uniformly, ignoring boundary vulnerabilities. In this paper, to improve classification performance, we tackle graph structural disentanglement by identifying boundary-region entanglement as the primary bottleneck and propose Boundary Embedding Shaping (BES), an adaptive contrastive learning GNN plug-in module that selectively suppresses spurious structural noise at decision boundaries with minimal model parameter perturbation. Extensive experiments demonstrate that BES consistently improves boundary discrimination and outperforms existing leading methods. Notably, BES boosts GCN performance by an average of 3.3% in node classification (up to 5.0% on WikiCS) and achieves superior accuracy in link prediction.

Problem

• graph structural entanglement
• boundary nodes
• spurious correlations
• decision boundary
• node classification

Innovation

• Boundary Embedding Shaping
• Adaptive Contrastive Learning
• Graph Structural Disentanglement
• Decision Boundary
• Spurious Correlation