Revisiting Positive Samples in Graph Contrastive Learning: From the Perspective of Message Passing

2026-06-08

AI Summary

This work addresses a critical yet overlooked issue in graph contrastive learning: the message-passing mechanism inadvertently attenuates the similarity signals of positive samples, leading to an overestimation of their learning efficacy. To uncover this degradation mechanism, the authors analyze the process through the lens of Dirichlet energy and propose SPGCL (Spectral-Preserved Graph Contrastive Learning). SPGCL decomposes node features into high- and low-energy components via energy-aware disentanglement; the low-energy component is used for message propagation to preserve structural smoothness, while the high-energy component enables reliable positive sample selection to retain discriminative signals. Extensive experiments across multiple graph benchmarks demonstrate that SPGCL consistently outperforms state-of-the-art methods, substantiating both the validity of the identified mechanism and the effectiveness of the proposed approach.

Abstract

Graph Contrastive Learning (GCL), which trains graph encoders by maximizing similarity between positive samples and minimizing it between negative ones, has emerged as a mainstream graph pre-training paradigm. It is widely recognized that positive samples are essential in GCLs. Ideally, maximizing the similarity of positive samples enables graph encoders to capture intrinsic semantic and patterns of graph data. However, we discover an interesting phenomenon: GCLs can achieve competitive performance even without positive samples. This motivates us to revisit the fundamental mechanism of positive samples in GCLs. From the perspective of Dirichlet energy, we theoretically finds that message passing, a key mechanism in graph encoders, trivializes the maximization of positive samples, preventing GCLs from effectively learning from positive samples. To address this, we propose SPGCL to mitigate the trivialization caused by message passing and restore the learning efficacy of positive samples. Specifically, we find that high Dirichlet energy features help positive samples provide effective learning signals while low Dirichlet energy features contribute little to positive learning signal but is useful for positive sampling. Based on this, SPGCL propagates only high Dirichlet energy features and uses low energy features to construct a probability matrix for reliable positive sampling. Extensive experiments demonstrate the effectiveness of SPGCL.

Problem

• Graph Contrastive Learning
• Positive Samples
• Message Passing
• Dirichlet Energy
• Representation Learning

Innovation

• Graph Contrastive Learning
• Message Passing
• Dirichlet Energy
• Positive Sampling
• SPGCL