Research map / Graph learning / Graph self-supervised learning
Contrastive learning with negatives and views
Improve graph contrastive learning via hard negatives, self-augmentation, and clustering.
51 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Graph self-supervised learning, keyword cluster “graph contrastive · gcl · augmentation”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 3 of 6 tracked benchmarks.
20 papers in 2023–24 and 20 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 5.8% to 3.4%.
Open questions the data points to
- Citeseer (accuracy): best went from 74.2 (up to 2024) to 77.1 in 2025 — LOHA: Direct Graph Spectral Contrastive Learning Between Low-Pass and High-Pass Views (AAAI 2025 · 7 citations)
- PubMed (accuracy): best went from 85.47 (up to 2024) to 87.3 in 2025 — LOHA: Direct Graph Spectral Contrastive Learning Between Low-Pass and High-Pass Views (AAAI 2025 · 7 citations)
- COLLAB (accuracy): best went from 81.76 (up to 2024) to 83 in 2025 — Graph Structure Refinement with Energy-based Contrastive Learning (AAAI 2025 · 3 citations)
- Papers report SimGCL on Gowalla (NDCG@20) anywhere from 0.0818 to 0.113: comparisons against it depend on whose number you take.
What looks settled
- Cora (accuracy): best up to 2024 84.6, best in 2025–26 85.1 from 2 papers, less than 1% better.
No longer reported by papers here: NCI1 (accuracy (%), last in 2024).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2018 · Contrastive Training for Models of Information Cascades (AAAI 2018 · 3 citations)
- 2021 · Contrastive Self-supervised Learning for Graph Classification (AAAI 2021 · 134 citations)
- 2022 · AutoGCL: Automated Graph Contrastive Learning via Learnable View Generators (AAAI 2022 · 152 citations)
- 2023 · Cluster-Guided Contrastive Graph Clustering Network (AAAI 2023 · 132 citations)
- 2024 · Improved Graph Contrastive Learning for Short Text Classification (AAAI 2024 · 24 citations)
- 2025 · BIG-FUSION: Brain-Inspired Global-Local Context Fusion Framework for Multimodal Emotion Recognition in Conversations (AAAI 2025 · 7 citations)
- 2026 · Stage-Aware Graph Contrastive Learning with Node-oriented Mixture of Experts (AAAI 2026 · 2 citations)
Earliest papers here
- Contrastive Training for Models of Information Cascades (AAAI 2018 · 3 citations)
- Contrastive Self-supervised Learning for Graph Classification (AAAI 2021 · 134 citations)
- Disentangled Contrastive Learning on Graphs (NeurIPS 2021 · 33 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
Citeseer · accuracy (higher is better)
9 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 73.5 | InfoGCL: Information-Aware Graph Contrastive Learning (NeurIPS 2021 · 32 citations) |
| 2022 | 74.2 | SAIL: Self-Augmented Graph Contrastive Learning (AAAI 2022 · 26 citations) |
| 2025 | 77.1 | LOHA: Direct Graph Spectral Contrastive Learning Between Low-Pass and High-Pass Views (AAAI 2025 · 7 citations) |
Cora · accuracy (higher is better)
8 papers, 2021–2025 · small gains only
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 83.5 | InfoGCL: Information-Aware Graph Contrastive Learning (NeurIPS 2021 · 32 citations) |
| 2022 | 84.6 | SAIL: Self-Augmented Graph Contrastive Learning (AAAI 2022 · 26 citations) |
| 2025 | 85.1 | LOHA: Direct Graph Spectral Contrastive Learning Between Low-Pass and High-Pass Views (AAAI 2025 · 7 citations) |
PubMed · accuracy (higher is better)
7 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 79.1 | InfoGCL: Information-Aware Graph Contrastive Learning (NeurIPS 2021 · 32 citations) |
| 2022 | 83.8 | SAIL: Self-Augmented Graph Contrastive Learning (AAAI 2022 · 26 citations) |
| 2024 | 85.47 | FUG: Feature-Universal Graph Contrastive Pre-training for Graphs with Diverse Node Features (NeurIPS 2024 · 2 citations) |
| 2025 | 87.3 | LOHA: Direct Graph Spectral Contrastive Learning Between Low-Pass and High-Pass Views (AAAI 2025 · 7 citations) |
COLLAB · accuracy (higher is better)
6 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 81.2 | Disentangled Contrastive Learning on Graphs (NeurIPS 2021 · 33 citations) |
| 2023 | 81.76 | Wiener Graph Deconvolutional Network Improves Graph Self-Supervised Learning (AAAI 2023 · 12 citations) |
| 2025 | 83 | Graph Structure Refinement with Energy-based Contrastive Learning (AAAI 2025 · 3 citations) |
NCI1 · accuracy (%) (higher is better)
6 papers, 2021–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 81.9 | Disentangled Contrastive Learning on Graphs (NeurIPS 2021 · 33 citations) |
| 2022 | 82 | AutoGCL: Automated Graph Contrastive Learning via Learnable View Generators (AAAI 2022 · 152 citations) |
PROTEINS · accuracy (%) (higher is better)
6 papers, 2021–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 82.77 | Contrastive Self-supervised Learning for Graph Classification (AAAI 2021 · 134 citations) |
What has been tried
38 papers here report 403 measured changes (a component added, removed or swapped, compared with a baseline): helped 67%, no clear effect 7%, hurt 21%, mixed 5%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add SVD-based contrastive learning and cluster-oriented contrastive learning | Twitter, MR, Snippets, StackOverflow · ACC | 69.60 -> 73.16; 61.52 -> 65.21; 79.25 -> 83.73; 76.20 -> 83.07 | Helped | Improved Graph Contrastive Learning for Short Text Classification (AAAI 2024 · 24 citations) |
| Use 1 GNN hop instead of 2 | Horse→Zebra · FID | 34.5 -> 37.9 | Hurt | Patch-Wise Graph Contrastive Learning for Image Translation (AAAI 2024 · 21 citations) |
| Add global objective L_glb | Cornell · NMI | 26.02 -> 37.51 | Helped | Sterling: Synergistic Representation Learning on Bipartite Graphs (AAAI 2024 · 18 citations) |
| add learnable dimensional rationale | ZINC-2M transfer benchmarks · average ROC-AUC | 71.9 -> 72.5 | Helped | Rethinking Dimensional Rationale in Graph Contrastive Learning from Causal Perspective (AAAI 2024 · 12 citations) |
| add topic and semantic review graphs to rating-graph modeling | Instrument, Music, and Toy · HR@10 | 0.2941 -> 0.3936; 0.4897 -> 0.6116; 0.3483 -> 0.3885 | Helped | Review-Enhanced Hierarchical Contrastive Learning for Recommendation (AAAI 2024 · 12 citations) |
| add GCL | IEMOCAP · WF1 | 68.50 -> 72.91 | Helped | BIG-FUSION: Brain-Inspired Global-Local Context Fusion Framework for Multimodal Emotion Recognition in Conversations (AAAI 2025 · 7 citations) |
| Use the LOHA-var-3 spectral signal trend variant instead of LOHA | Nine real-world datasets · accuracy | Hurt | LOHA: Direct Graph Spectral Contrastive Learning Between Low-Pass and High-Pass Views (AAAI 2025 · 7 citations) | |
| add the Word-POS contrastive view | Twitter, MR, Snippets, StackOverflow · ACC | 70.60 -> 72.59; 61.94 -> 64.18; 79.27 -> 83.63; 80.29 -> 82.26 | Helped | A Simple Graph Contrastive Learning Framework for Short Text Classification (AAAI 2025 · 6 citations) |
| Remove structural embeddings and use raw graph attributes | Cora, Citeseer, Actor, and Pubmed · accuracy | Hurt | Graph Structure Refinement with Energy-based Contrastive Learning (AAAI 2025 · 3 citations) | |
| Add street view images as geographic-configuration inputs | Singapore road function prediction · Macro-F1 (%) | 30.67 ± 0.66 -> 50.21 ± 0.60 | Helped | Road Network Representation Learning with the Third Law of Geography (NeurIPS 2024 · 3 citations) |
| Add LDE dimensional-encoding loss | Photo · accuracy | 57.37 ± 3.38 -> 93.07 ± 0.82 | Helped | FUG: Feature-Universal Graph Contrastive Pre-training for Graphs with Diverse Node Features (NeurIPS 2024 · 2 citations) |
| Add graph contrastive learning (GCL) to GSCExplainer | MUTAG · prediction accuracy | drops of 6.4 percentage points | Helped | Graph Segmentation and Contrastive Enhanced Explainer for Graph Neural Networks (AAAI 2025 · 0 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| SimGCL | Gowalla | NDCG@20 | 0.0818 – 0.113 | 2 papers, e.g. ICLR 2023 and NeurIPS 2023 |
What reviewers will expect
Datasets and baselines named in the 11 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- MUTAG · 2 of 11 papers
- Cora · 2 of 11 papers
- Citeseer · 2 of 11 papers
- Pubmed · 2 of 11 papers
- Cornell · 2 of 11 papers
- Texas · 2 of 11 papers
- Wisconsin · 2 of 11 papers
- Actor · 2 of 11 papers
Baselines
- GCN · 3 of 11 papers
- GraphCL · 2 of 11 papers
- GAT · 2 of 11 papers
- DGI · 2 of 11 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 40 papers; from OpenAlex and the papers' affiliations).
- Beihang University · 4 papers
- The University of Hong Kong · 3 papers
- University of Science and Technology of China · 2 papers
- Nanjing University of Aeronautics and Astronautics · 2 papers
- University of Trento · 2 papers
- Jilin University · 2 papers
- Korea Advanced Institute of Science and Technology · 2 papers
- University of Chinese Academy of Sciences · 2 papers
- Tsinghua University · 2 papers
- Shanxi University · 2 papers
Use it
Download BibTeX of 33 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Graph contrastive learning analysis (61 papers)
- Augmentation-free graph learning (25 papers)
Elsewhere in the topic
- Semi-supervised node representation learning (214 papers)
- Heterophily and imbalance on graphs (178 papers)
- Diffusion-based graph generation (138 papers)
- LLMs and graph foundation models (117 papers)
- Graph anomaly detection (101 papers)
- Molecular graph pretraining (101 papers)
- Graph domain adaptation (82 papers)
- Brain and biological graph learning (78 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).