Scholarlink

<p>Graph Contrastive Learning (GCL) has made significant progress in capturing representations from unlabeled graphs. However, existing methods often fail to accurately identify true positive and negative node pairs, limiting their effectiveness. To address this, we propose Positive Mining Graph Contrastive Learning (PMGCL), a novel approach that leverages a Beta Mixture Model (BMM) to estimate the probability of true positive pairs between nodes. Unlike traditional GCL methods that rely on a single positive pair or heuristic rules, PMGCL allows for multiple positive pairs per anchor node, significantly improving the accuracy of positive sample selection. We have conducted a comprehensive evaluation of PMGCL on a range of real-world graph datasets. The experimental findings indicate that PMGCL significantly outperforms traditional GCL methods. Our method not only achieves excellent results in unsupervised learning benchmarks but also exceeds the performance of supervised learning benchmarks in certain scenarios.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Positive mining in graph contrastive learning

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Guodong Zheng,
  • Yongli Cheng

摘要

Graph Contrastive Learning (GCL) has made significant progress in capturing representations from unlabeled graphs. However, existing methods often fail to accurately identify true positive and negative node pairs, limiting their effectiveness. To address this, we propose Positive Mining Graph Contrastive Learning (PMGCL), a novel approach that leverages a Beta Mixture Model (BMM) to estimate the probability of true positive pairs between nodes. Unlike traditional GCL methods that rely on a single positive pair or heuristic rules, PMGCL allows for multiple positive pairs per anchor node, significantly improving the accuracy of positive sample selection. We have conducted a comprehensive evaluation of PMGCL on a range of real-world graph datasets. The experimental findings indicate that PMGCL significantly outperforms traditional GCL methods. Our method not only achieves excellent results in unsupervised learning benchmarks but also exceeds the performance of supervised learning benchmarks in certain scenarios.