Research map / Graph learning / Graph self-supervised learning
Heterophily and imbalance on graphs
Handle heterophilic, imbalanced, and heterogeneous graphs in node learning.
178 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Graph self-supervised learning, keyword cluster “classification · semi supervised · representation”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 6 tracked benchmarks.
51 papers in 2023–24 and 70 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 14.8% to 12.0%.
Open questions the data points to
- Cora (accuracy): best went from 83.9 (up to 2024) to 87.25 in 2025 — Self-supervised Masked Graph Autoencoder via Structure-aware Curriculum (ICML 2025)
What looks settled
- Citeseer (accuracy): best up to 2024 74.2, best in 2025–26 74.68 from 2 papers, less than 1% better.
- Pubmed (accuracy): best up to 2024 90.08, best in 2025–26 85.86 from 2 papers, at most 0.2% better, if at all.
- Coauthor CS (accuracy): best up to 2024 95.4, best in 2025–26 92.69 from 2 papers, at most 0.2% better, if at all.
- Coauthor Physics (accuracy): best up to 2024 97.3, best in 2025–26 95.91 from 2 papers, at most 0.2% better, if at all.
No longer reported by papers here: MUTAG (accuracy, last in 2022).
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
- 2017 · Community Preserving Network Embedding (AAAI 2017 · 933 citations)
- 2018 · Beyond Link Prediction: Predicting Hyperlinks in Adjacency Space (AAAI 2018 · 106 citations)
- 2019 · Label Efficient Semi-Supervised Learning via Graph Filtering (CVPR 2019 · 173 citations)
- 2020 · Unsupervised Attributed Multiplex Network Embedding (AAAI 2020 · 283 citations)
- 2021 · Subgraph Federated Learning with Missing Neighbor Generation (NeurIPS 2021 · 76 citations)
- 2022 · Node-Aligned Graph Convolutional Network for Whole-Slide Image Representation and Classification (CVPR 2022 · 68 citations)
- 2023 · Heterogeneous Graph Masked Autoencoders (AAAI 2023 · 107 citations)
- 2024 · Structural Entropy Based Graph Structure Learning for Node Classification (AAAI 2024 · 23 citations)
- 2025 · Revolutionizing Encrypted Traffic Classification with MH-Net: A Multi-View Heterogeneous Graph Model (AAAI 2025 · 30 citations)
- 2026 · Relational Graph Transformer (ICLR 2026 · 1 citation)
Earliest papers here
- Community Preserving Network Embedding (AAAI 2017 · 933 citations)
- Scalable Graph Embedding for Asymmetric Proximity (AAAI 2017 · 209 citations)
- Deep Learning with Topological Signatures (NeurIPS 2017 · 100 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, 2020–2025 · small gains only
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 70.7 | Uncertainty Aware Graph Gaussian Process for Semi-Supervised Learning (AAAI 2020 · 21 citations) |
| 2022 | 72.14 | Automated Self-Supervised Learning for Graphs (ICLR 2022 · 21 citations) |
| 2023 | 73.08 | Beyond Smoothing: Unsupervised Graph Representation Learning with Edge Heterophily Discriminating (AAAI 2023 · 62 citations) |
| 2024 | 74.2 | Motif-Aware Riemannian Graph Neural Network with Generative-Contrastive Learning (AAAI 2024 · 21 citations) |
| 2025 | 74.68 | Self-supervised Masked Graph Autoencoder via Structure-aware Curriculum (ICML 2025) |
Pubmed · accuracy (higher is better)
8 papers, 2020–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 76.7 | Uncertainty Aware Graph Gaussian Process for Semi-Supervised Learning (AAAI 2020 · 21 citations) |
| 2022 | 90.08 | Why Propagate Alone? Parallel Use of Labels and Features on Graphs (ICLR 2022 · 5 citations) |
Cora · accuracy (higher is better)
7 papers, 2020–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 82.7 | Uncertainty Aware Graph Gaussian Process for Semi-Supervised Learning (AAAI 2020 · 21 citations) |
| 2023 | 83.81 | Beyond Smoothing: Unsupervised Graph Representation Learning with Edge Heterophily Discriminating (AAAI 2023 · 62 citations) |
| 2024 | 83.9 | Enhancing Graph Transformers with Hierarchical Distance Structural Encoding (NeurIPS 2024 · 2 citations) |
| 2025 | 87.25 | Self-supervised Masked Graph Autoencoder via Structure-aware Curriculum (ICML 2025) |
Coauthor CS · accuracy (higher is better)
5 papers, 2022–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 95.4 | Omni-Granular Ego-Semantic Propagation for Self-Supervised Graph Representation Learning (ICML 2022 · 1 citation) |
Coauthor Physics · accuracy (higher is better)
5 papers, 2022–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 97.3 | Omni-Granular Ego-Semantic Propagation for Self-Supervised Graph Representation Learning (ICML 2022 · 1 citation) |
MUTAG · accuracy (higher is better)
5 papers, 2019–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 88.3 | Learning metrics for persistence-based summaries and applications for graph classification (NeurIPS 2019 · 54 citations) |
| 2022 | 95.3 | Omni-Granular Ego-Semantic Propagation for Self-Supervised Graph Representation Learning (ICML 2022 · 1 citation) |
What has been tried
130 papers here report 1,287 measured changes (a component added, removed or swapped, compared with a baseline): helped 73%, no clear effect 7%, hurt 16%, 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 the 4-bit view | CIC-IoT flow-level · F1 | 0.9800 -> 0.9896 | Helped | Revolutionizing Encrypted Traffic Classification with MH-Net: A Multi-View Heterogeneous Graph Model (AAAI 2025 · 30 citations) |
| replace community-influence plus confidence fusion with average fusion | Cancer, Polblogs, Citeseer, Wiki-CS · F1-micro (%) | 95.8 -> 95.0; 95.9 -> 95.6; 74.1 -> 73.4; 75.6 -> 74.9 | Hurt | Structural Entropy Based Graph Structure Learning for Node Classification (AAAI 2024 · 23 citations) |
| add motif-aware hardness to contrastive learning | Cora and Citeseer · AUC | 97.05 -> 97.37; 98.36 -> 98.54 | Helped | Motif-Aware Riemannian Graph Neural Network with Generative-Contrastive Learning (AAAI 2024 · 21 citations) |
| Add adaptive HIN selection | CiteSeer feature unlearning with GAT · F1 Score | 72.8±0.3 -> 74.3±0.2 | Helped | Towards Effective and General Graph Unlearning via Mutual Evolution (AAAI 2024 · 17 citations) |
| Add dual-level supervision | Cora · ACC | 70.75 -> 72.19 | Helped | Every Node Is Different: Dynamically Fusing Self-Supervised Tasks for Attributed Graph Clustering (AAAI 2024 · 15 citations) |
| model node-label dependencies with GLDL instead of GLDLs | DBLP · KLD | 0.3455 -> 0.3449 | No clear effect | GLDL: Graph Label Distribution Learning (AAAI 2024 · 11 citations) |
| add a cross-GNN layer rather than using a GCN-form layer | Facebook/Twitter (Facebook layer), 90% interlayer links · AUC | 0.8967 -> 0.9653 | Helped | Link Prediction in Multilayer Networks via Cross-Network Embedding (AAAI 2024 · 11 citations) |
| Add structural features to the feature matrix (w/o FM → FairGP) | Pokec-z-R, Pokec-z-G, AMiner-L · ΔSP (%) | 1.07 -> 0.33; 3.14 -> 1.18; 5.05 -> 4.44 | Helped | FairGP: A Scalable and Fair Graph Transformer Using Graph Partitioning (AAAI 2025 · 7 citations) |
| Add counterfactual item links | Ciao and Epinions · HR@N | 0.6203 -> 0.6586; 0.3993 -> 0.4120 | Helped | Social Recommendation via Graph-Level Counterfactual Augmentation (AAAI 2025 · 7 citations) |
| Replace subgraph-centric mixup with node-centric mixup (r=0) | Pubmed · accuracy | Hurt | AGMixup: Adaptive Graph Mixup for Semi-supervised Node Classification (AAAI 2025 · 6 citations) | |
| add the complete prototype graph instead of using an isolated graph | Citeseer · F1 | 65.02 -> 66.42 | Helped | THESAURUS: Contrastive Graph Clustering by Swapping Fused Gromov-Wasserstein Couplings (AAAI 2025 · 6 citations) |
| Add shared-information loss $\mathcal{L}_s$ | ACM, DBLP, Yelp · Macro-F1 | 93.05±0.49 -> 93.42±0.21 (ACM); 90.44±0.45 -> 91.28±0.31 (DBLP); 93.15±0.12 -> 93.26±0.26 (Yelp) | Helped | Beyond Redundancy: Information-aware Unsupervised Multiplex Graph Structure Learning (NeurIPS 2024 · 6 citations) |
What reviewers will expect
Datasets and baselines named in the 47 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 · 12 of 47 papers
- PubMed · 12 of 47 papers
- Citeseer · 10 of 47 papers
- ogbn-arxiv · 7 of 47 papers
- Amazon-Photo · 5 of 47 papers
- Coauthor-Physics · 4 of 47 papers
- DBLP · 4 of 47 papers
- Coauthor CS · 3 of 47 papers
Baselines
- GCN · 11 of 47 papers
- GAT · 6 of 47 papers
- FedAvg · 4 of 47 papers
- FedProx · 4 of 47 papers
- FedSage+ · 4 of 47 papers
- GIN · 4 of 47 papers
- FairGNN · 3 of 47 papers
- HAN · 3 of 47 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 121 papers; from OpenAlex and the papers' affiliations).
- University of Electronic Science and Technology of China · 7 papers
- Beihang University · 7 papers
- Guangxi Normal University · 5 papers
- National University of Singapore · 5 papers
- Griffith University · 4 papers
- The Hong Kong Polytechnic University · 4 papers
- Xidian University · 4 papers
- Wuhan University · 4 papers
- Monash University · 3 papers
- Southeast University · 3 papers
Use it
Download BibTeX of 40 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
- Semi-supervised node representation learning (214 papers)
- LLMs and graph foundation models (117 papers)
- Brain and biological graph learning (78 papers)
- Multi-view graph clustering (70 papers)
Elsewhere in the topic
- Diffusion-based graph generation (138 papers)
- Graph anomaly detection (101 papers)
- Molecular graph pretraining (101 papers)
- Graph domain adaptation (82 papers)
- Graph contrastive learning analysis (61 papers)
- Graph out-of-distribution generalization (58 papers)
- Contrastive learning with negatives and views (51 papers)
- Augmentation-free graph learning (25 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).