Research map / Graph learning / Graph self-supervised learning
Eliciting Structural and Semantic Global Knowledge in Unsupervised Graph Contrastive Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use S³-CL rather than the strongest self-supervised baseline | node classification · Cora, Citeseer, Amazon-P, Coauthor CS | accuracy: 83.4\pm{0.5} -> 84.5\pm{0.4}; 74.0\pm{0.7} -> 74.6\pm{0.4}; 88.8\pm0.4 -> 89.0\pm{0.5}; 92.4\pm{0.4} -> 93.1\pm{0.4} | Helped |
| Use S³-CL rather than SUGRL | node classification · Pubmed | accuracy: 81.9\pm\textbf{0.5} -> 80.8\pm0.3 | Hurt |
| Use S³-CL rather than GAT | node classification · ogbn-arxiv | accuracy: 73.2\pm0.2 -> 72.8\pm0.3 | Hurt |
| Use S³-CL rather than the strongest clustering baseline | node clustering · Cora, Citeseer, Pubmed | NMI: 58.5 -> 60.7; 45.8 -> 46.3; 35.0 -> 36.0 | Helped |
| Add structural contrastive learning (ablation) | S³-CL · Citeseer, Cora, Pubmed | accuracy: 73.1$\pm{0.2}$ -> 74.6$\pm{0.4}$; 83.3$\pm{0.3}$ -> 84.5$\pm{0.4}$; 80.0$\pm$ 0.3 -> 80.8$\pm{0.3}$ | Helped |
| Add semantic contrastive learning (ablation) | S³-CL · Citeseer, Cora, Pubmed | accuracy: 71.9$\pm{0.4}$ -> 74.6$\pm{0.4}$; 82.2$\pm{0.5}$ -> 84.5$\pm{0.4}$; 79.3$\pm$ 0.2 -> 80.8$\pm{0.3}$ | Helped |
| Replace Bayesian non-parametric prototype inference with K-Means (ablation) | S³-CL · Cora, Citeseer, Pubmed | accuracy: 84.8$\pm{0.6}$ vs 84.5$\pm{0.4}$; 74.6$\pm{0.3}$ vs 74.6$\pm{0.4}$; 80.7$\pm{0.2}$ vs 80.8$\pm{0.3}$ | Mixed |
| Increase structural-loss negative samples from 64 to 512 (ablation) | S³-CL · Cora, Citeseer, Pubmed | accuracy: 84.0$\pm$0.2 -> 84.5$\pm0.4; 74.1$\pm$ 0.3 -> 74.6$\pm$ 0.4; 80.3$\pm$0.3 -> 80.8$\pm$0.3 | Helped |
| Increase structural-loss negative samples beyond 512 (ablation) | S³-CL · Cora, Citeseer, Pubmed | accuracy: 84.5$\pm0.4; 74.6$\pm$ 0.4; 80.8$\pm$0.3 -> 84.6$\pm0.7, 74.6$\pm$ 0.6, 80.7$\pm$0.6 (1024); 84.6$\pm0.5, 74.5$\pm$ 0.2, | No clear effect |
Closest papers by meaning
- ReGCL: Rethinking Message Passing in Graph Contrastive Learning (AAAI 2024 · similarity 0.92)
- Does GCL Need a Large Number of Negative Samples? Enhancing Graph Contrastive Learning with Effective and Efficient Negative Sampling (AAAI 2025 · similarity 0.91)
- S3GCL: Spectral, Swift, Spatial Graph Contrastive Learning (ICML 2024 · similarity 0.91)
- Beyond Homophily: Graph Contrastive Learning with Macro-Micro Message Passing (AAAI 2025 · similarity 0.91)
- TopoGCL: Topological Graph Contrastive Learning (AAAI 2024 · similarity 0.90)
- Graph Structure Refinement with Energy-based Contrastive Learning (AAAI 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.