Research map / Label-efficient and robust learning / Semi-supervised learning
I’m Me, We’re Us, and I’m Us: Tri-directional Contrastive Learning on Hypergraphs
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 |
|---|---|---|---|
| add group-level contrast to node-level contrast (ablation) | TriCL · Citeseer | accuracy: 70.28 $\pm$ 1.5 -> 71.38 $\pm$ 1.4 | Helped |
| add membership-level contrast to node- and group-level contrast (ablation) | TriCL · Citeseer | accuracy: 71.38 $\pm$ 1.4 -> 72.02 $\pm$ 1.4 | Helped |
| use membership-level contrast alone instead of node-level contrast (ablation) | TriCL · DBLP | accuracy: 90.88 $\pm$ 0.1 -> 63.95 $\pm$ 7.2 | Hurt |
| add self-loops to the input hypergraph (ablation) | TriCL · Cora-A | accuracy: 79.77 $\pm$ 0.95 -> 82.15 $\pm$ 0.89 | Helped |
| add self-loops to the input hypergraph (ablation) | TriCL · Zoo | accuracy: 80.43 $\pm$ 11.3 -> 80.25 $\pm$ 11.2 | Hurt |
| replace the mean pooling encoder with HGNN (ablation) | TriCL · Cora-C | accuracy: 81.57 $\pm$ 1.12 -> 81.03 $\pm$ 1.31 | Hurt |
| subsample two rather than all negatives (ablation) | TriCL · Cora-C | accuracy: 81.57 $\pm$ 1.1 -> 80.62 $\pm$ 1.3 | Hurt |
| use two sampled negatives rather than all negatives (ablation) | TriCL · average performance degradation | A.P.D.: 0.49% | No clear effect |
| combine structural membership masking with node feature masking instead of using only one augmentation (ablation) | TriCL · node classification | accuracy | Helped |
| add node-, group-, and membership-level contrast together (ablation) | TriCL · node classification across datasets | accuracy | Helped |
Closest papers by meaning
- Heterophily-aware Contrastive Learning for Heterophilic Hypergraphs (AAAI 2026 · similarity 0.91)
- Does GCL Need a Large Number of Negative Samples? Enhancing Graph Contrastive Learning with Effective and Efficient Negative Sampling (AAAI 2025 · similarity 0.89)
- Hierarchical Topology Isomorphism Expertise Embedded Graph Contrastive Learning (AAAI 2024 · similarity 0.88)
- Perfect Alignment May be Poisonous to Graph Contrastive Learning (ICML 2024 · similarity 0.88)
- ReGCL: Rethinking Message Passing in Graph Contrastive Learning (AAAI 2024 · similarity 0.88)
- SOLA-GCL: Subgraph-Oriented Learnable Augmentation Method for Graph Contrastive Learning (AAAI 2025 · similarity 0.88)
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.