Research map / Graph learning / Graph self-supervised learning
Graph contrastive learning analysis
Study and scale graph contrastive learning with theory and efficient objectives.
61 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2020–2026) · 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 1 of 6 tracked benchmarks.
25 papers in 2023–24 and 26 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 7.3% to 4.5%.
Open questions the data points to
- Cornell (accuracy): best went from 81.27 (up to 2024) to 85.03 in 2025 — Beyond Homophily: Graph Contrastive Learning with Macro-Micro Message Passing (AAAI 2025 · 2 citations)
- Papers report GraphCL on NCI1 (accuracy) anywhere from 68.54 to 77.87: comparisons against it depend on whose number you take.
What looks settled
- PubMed (accuracy): best up to 2024 87.89, best in 2025–26 86.72 from 3 papers, at most 0.2% better, if at all.
- Citeseer (accuracy): best up to 2024 76.31, best in 2025–26 73.29 from 2 papers, at most 0.2% better, if at all.
- Cora (accuracy): best up to 2024 88.47, best in 2025–26 85.62 from 3 papers, at most 0.2% better, if at all.
- Wiki-CS (accuracy): best up to 2024 82.13, best in 2025–26 82.09 from 4 papers, at most 0.2% better, if at all.
- Actor (accuracy): best up to 2024 36.88, best in 2025–26 35.48 from 2 papers, at most 0.2% better, if at all.
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
- 2020 · Graph Contrastive Learning with Augmentations (NeurIPS 2020 · 859 citations)
- 2021 · Adversarial Graph Augmentation to Improve Graph Contrastive Learning (NeurIPS 2021 · 141 citations)
- 2022 · Rethinking and Scaling Up Graph Contrastive Learning: An Extremely Efficient Approach with Group Discrimination (NeurIPS 2022 · 50 citations)
- 2023 · Neighbor Contrastive Learning on Learnable Graph Augmentation (AAAI 2023 · 121 citations)
- 2024 · TopoGCL: Topological Graph Contrastive Learning (AAAI 2024 · 29 citations)
- 2025 · Why Does Dropping Edges Usually Outperform Adding Edges in Graph Contrastive Learning? (AAAI 2025 · 18 citations)
- 2026 · Dual-Kernel Graph Community Contrastive Learning (AAAI 2026 · 1 citation)
Earliest papers here
- Graph Contrastive Learning with Augmentations (NeurIPS 2020 · 859 citations)
- Adversarial Graph Augmentation to Improve Graph Contrastive Learning (NeurIPS 2021 · 141 citations)
- Graph Contrastive Learning Automated (ICML 2021 · 93 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.
PubMed · accuracy (higher is better)
14 papers, 2021–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 80.69 | Directed Graph Contrastive Learning (NeurIPS 2021 · 18 citations) |
| 2023 | 86.29 | Contrastive Learning Meets Homophily: Two Birds with One Stone (ICML 2023) |
| 2024 | 87.89 | S3GCL: Spectral, Swift, Spatial Graph Contrastive Learning (ICML 2024) |
Citeseer · accuracy (higher is better)
13 papers, 2021–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 67.42 | Directed Graph Contrastive Learning (NeurIPS 2021 · 18 citations) |
| 2023 | 74.6 | Eliciting Structural and Semantic Global Knowledge in Unsupervised Graph Contrastive Learning (AAAI 2023 · 30 citations) |
| 2024 | 76.31 | S3GCL: Spectral, Swift, Spatial Graph Contrastive Learning (ICML 2024) |
Cora · accuracy (higher is better)
13 papers, 2023–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 85.86 | Spectral Augmentation for Self-Supervised Learning on Graphs (ICLR 2023 · 16 citations) |
| 2024 | 88.47 | S3GCL: Spectral, Swift, Spatial Graph Contrastive Learning (ICML 2024) |
Wiki-CS · accuracy (higher is better)
8 papers, 2023–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 82.13 | Spectral Augmentation for Self-Supervised Learning on Graphs (ICLR 2023 · 16 citations) |
Actor · accuracy (higher is better)
7 papers, 2023–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 33.6 | Attribute and Structure Preserving Graph Contrastive Learning (AAAI 2023 · 55 citations) |
| 2024 | 36.88 | S3GCL: Spectral, Swift, Spatial Graph Contrastive Learning (ICML 2024) |
Cornell · accuracy (higher is better)
7 papers, 2023–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 78.2 | Attribute and Structure Preserving Graph Contrastive Learning (AAAI 2023 · 55 citations) |
| 2024 | 81.27 | S3GCL: Spectral, Swift, Spatial Graph Contrastive Learning (ICML 2024) |
| 2025 | 85.03 | Beyond Homophily: Graph Contrastive Learning with Macro-Micro Message Passing (AAAI 2025 · 2 citations) |
What has been tried
48 papers here report 525 measured changes (a component added, removed or swapped, compared with a baseline): helped 68%, no clear effect 6%, hurt 21%, mixed 6%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| replace EPI with PI | MUTAG · accuracy | 89.78±1.33 -> 90.09±0.93 | Helped | TopoGCL: Topological Graph Contrastive Learning (AAAI 2024 · 29 citations) |
| add the independence objective | Cora · Ma-F1 | 57.5 -> 83.8 | Helped | Graph Contrastive Invariant Learning from the Causal Perspective (AAAI 2024 · 20 citations) |
| Remove gradient-guided structure learning (GGSL) | node classification benchmarks · accuracy | Hurt | ReGCL: Rethinking Message Passing in Graph Contrastive Learning (AAAI 2024 · 20 citations) | |
| add feature masking | Cora, CiteSeer, PubMed, WikiCS, Amazon-Photo, Coauthor-Physics, ogbn-arxiv · accuracy | 85.04 ± 0.33 -> 86.07 ± 0.32; 71.97 ± 0.62 -> 71.94 ± 0.57; 86.72 ± 0.11 -> 86.77 ± 0.14; 81.81 ± 0.18 -> 81.19 ± 0.11; | Mixed | Why Does Dropping Edges Usually Outperform Adding Edges in Graph Contrastive Learning? (AAAI 2025 · 18 citations) |
| add adversarial masking to GraphMAE | Cora · accuracy | 84.04 ± 0.58 -> 84.30 ± 0.75 | Helped | Rethinking Graph Masked Autoencoders through Alignment and Uniformity (AAAI 2024 · 18 citations) |
| Add graph-tier topology isomorphism expertise | NCI1 and PROTEINS · accuracy | Helped | Hierarchical Topology Isomorphism Expertise Embedded Graph Contrastive Learning (AAAI 2024 · 13 citations) | |
| add neighbor pruning and head-to-forged-tail alignment | Tail nodes across six datasets · AUC | 2.46% | Helped | Revisiting Graph Contrastive Learning on Anomaly Detection: A Structural Imbalance Perspective (AAAI 2025 · 10 citations) |
| Add RSM and TCM to the SGRL encoder | WikiCS · F1-score | 76.86 ± 0.74 -> 79.40 ± 0.13 | Helped | Exploitation of a Latent Mechanism in Graph Contrastive Learning: Representation Scattering (NeurIPS 2024 · 9 citations) |
| replace representative-node sampling with random sampling | CS, Photo, Computers, Wiki-CS, and Physics node classification · accuracy | 92.87 ± 0.47 -> 93.48 ± 0.59; 92.42 ± 1.31 -> 93.36 ± 0.76; 87.83 ± 0.70 -> 88.72 ± 0.96; 79.80 ± 1.22 -> 80.89 ± 1.21; | Helped | Does GCL Need a Large Number of Negative Samples? Enhancing Graph Contrastive Learning with Effective and Efficient Negative Sampling (AAAI 2025 · 6 citations) |
| use edge contrastive learning instead of the w/o ECL loss variant | Cora, 1 label/class · accuracy | 48.1 -> 63.6 | Helped | Edge Contrastive Learning: An Augmentation-Free Graph Contrastive Learning Model (AAAI 2025 · 5 citations) |
| Add dropout to the encoder | Transductive node classification (five datasets) · accuracy | 80.04±0.55 vs 80.91±0.46 (Wiki CS); 93.85±0.26 vs 95.32±0.19 (Amazon Photo); 89.72±0.36 vs 90.46± | Helped | Single-View Graph Contrastive Learning with Soft Neighborhood Awareness (AAAI 2025 · 5 citations) |
| use GN instead of Louvain partitioning | MUTAG · accuracy | 90.49±2.22 -> 89.88±2.98 | Hurt | SOLA-GCL: Subgraph-Oriented Learnable Augmentation Method for Graph Contrastive Learning (AAAI 2025 · 4 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 |
|---|---|---|---|---|
| MVGRL | Cora | accuracy (%) | 82.4 – 87.67 | 4 papers, e.g. AAAI 2025 and ICML 2024 |
| GraphCL | NCI1 | accuracy | 68.54 – 77.87 | 3 papers, e.g. NeurIPS 2021 and ICML 2021 |
| MVGRL | PubMed | accuracy (%) | 79.98 – 86.98 | 3 papers, e.g. NeurIPS 2021 and ICML 2024 |
| GRACE* | Wisconsin | accuracy | 49.02 – 77.5 | 2 papers, e.g. ICML 2023 and AAAI 2023 |
| GRACE | Wiki-CS | accuracy | 54.2 – 80.14 | 2 papers, e.g. AAAI 2023 and ICLR 2023 |
| GRACE* | Texas | accuracy | 55.86 – 74.8 | 2 papers, e.g. ICML 2023 and AAAI 2023 |
What reviewers will expect
Datasets and baselines named in the 17 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Cora · 9 of 17 papers
- PubMed · 9 of 17 papers
- CiteSeer · 8 of 17 papers
- Amazon-Photo · 7 of 17 papers
- Wiki-CS · 6 of 17 papers
- Coauthor-CS · 5 of 17 papers
- Amazon-Computers · 5 of 17 papers
- Cornell · 3 of 17 papers
Baselines
- GCA · 9 of 17 papers
- MVGRL · 8 of 17 papers
- GCN · 8 of 17 papers
- DGI · 8 of 17 papers
- GRACE · 8 of 17 papers
- GMI · 6 of 17 papers
- AFGRL · 5 of 17 papers
- BGRL · 5 of 17 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 51 papers; from OpenAlex and the papers' affiliations).
- Northwestern Polytechnical University · 8 papers
- Beihang University · 6 papers
- Tsinghua University · 5 papers
- Beijing University of Posts and Telecommunications · 5 papers
- Tianjin University · 5 papers
- George Washington University · 3 papers
- Hebei University of Technology · 3 papers
- Pennsylvania State University · 3 papers
- University of Chinese Academy of Sciences · 2 papers
- Chinese Academy of Sciences · 2 papers
Use it
Download BibTeX of 34 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
- Contrastive learning with negatives and views (51 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).