Boundary Embedding Shaping with Adaptive Contrastive Learning for Graph Structural Disentanglement

The paper introduces Boundary Embedding Shaping (BES), an adaptive contrastive learning module designed to mitigate graph structural entanglement in Graph Neural Networks (GNNs) by selectively suppressing spurious noise at class boundaries. Experimental results show that BES improves node classification performance by an average of 3.3% for Graph Convolutional Networks (GCNs), with a peak improvement of 5.0% on the WikiCS dataset, and enhances accuracy in link prediction tasks. This approach is significant for practitioners as it addresses boundary vulnerabilities in GNNs without extensive modifications to model parameters, potentially leading to more robust classification in complex graph structures.