Research map / Graph learning / Graph self-supervised learning
Graph anomaly detection
Detect anomalous nodes and graphs with generalist, generative, and diffusion-based methods.
101 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Graph self-supervised learning, keyword cluster “detection · anomaly · ood”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Leveling off Several papers from 2025–26 still report on 2 tracked benchmarks, but the best result has barely moved since 2024.
27 papers in 2023–24 and 55 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 7.8% to 9.4%.
What looks settled
- Amazon (AUPRC): best up to 2024 0.792, best in 2025–26 0.788 from 2 papers, at most 0.2% better, if at all.
- Reddit (AUPRC): best up to 2024 0.061, best in 2025–26 0.0616 from 2 papers, less than 1% better.
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
- 2016 · Efficient Nonparametric Subgraph Detection Using Tree Shaped Priors (AAAI 2016 · 13 citations)
- 2017 · Multiple Source Detection without Knowing the Underlying Propagation Model (AAAI 2017 · 95 citations)
- 2018 · Robust Detection of Link Communities in Large Social Networks by Exploiting Link Semantics (AAAI 2018 · 23 citations)
- 2019 · Community Detection in Social Networks Considering Topic Correlations (AAAI 2019 · 22 citations)
- 2020 · Rumor Detection on Social Media with Bi-Directional Graph Convolutional Networks (AAAI 2020 · 748 citations)
- 2021 · Towards Consumer Loan Fraud Detection: Graph Neural Networks with Role-Constrained Conditional Random Field (AAAI 2021 · 75 citations)
- 2022 · LUNAR: Unifying Local Outlier Detection Methods via Graph Neural Networks (AAAI 2022 · 130 citations)
- 2023 · Graph Anomaly Detection via Multi-Scale Contrastive Learning Networks with Augmented View (AAAI 2023 · 134 citations)
- 2024 · Revisiting Graph-Based Fraud Detection in Sight of Heterophily and Spectrum (AAAI 2024 · 57 citations)
- 2025 · Dynamic Spectral Graph Anomaly Detection (AAAI 2025 · 25 citations)
- 2026 · Correcting False Alarms from Unseen: Adapting Graph Anomaly Detectors at Test Time (AAAI 2026 · 5 citations)
Earliest papers here
- Efficient Nonparametric Subgraph Detection Using Tree Shaped Priors (AAAI 2016 · 13 citations)
- Multiple Source Detection without Knowing the Underlying Propagation Model (AAAI 2017 · 95 citations)
- Robust Detection of Link Communities in Large Social Networks by Exploiting Link Semantics (AAAI 2018 · 23 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.
Amazon · AUPRC (higher is better)
4 papers, 2023–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 0.263 | Truncated Affinity Maximization: One-class Homophily Modeling for Graph Anomaly Detection (NeurIPS 2023 · 9 citations) |
| 2024 | 0.792 | Generative Semi-supervised Graph Anomaly Detection (NeurIPS 2024 · 5 citations) |
Reddit · AUPRC (higher is better)
4 papers, 2023–2025 · small gains only
| Year | Best so far | Set by |
|---|---|---|
| 2023 | 0.0446 | Truncated Affinity Maximization: One-class Homophily Modeling for Graph Anomaly Detection (NeurIPS 2023 · 9 citations) |
| 2024 | 0.061 | Generative Semi-supervised Graph Anomaly Detection (NeurIPS 2024 · 5 citations) |
| 2025 | 0.0616 | Semi-supervised Graph Anomaly Detection via Robust Homophily Learning (NeurIPS 2025 · 1 citation) |
What has been tried
65 papers here report 616 measured changes (a component added, removed or swapped, compared with a baseline): helped 70%, no clear effect 6%, hurt 18%, 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 |
|---|---|---|---|---|
| add low-order pure high-pass bands | Amazon and T-Finance · F1-macro | 1.27% | Helped | Revisiting Graph-Based Fraud Detection in Sight of Heterophily and Spectrum (AAAI 2024 · 57 citations) |
| use the combined node-, edge-, and subgraph-level branches rather than random augmentation | Cora · AUC (%) | 81.61±0.01 -> 84.73±0.01 | Helped | ADA-GAD: Anomaly-Denoised Autoencoders for Graph Anomaly Detection (AAAI 2024 · 52 citations) |
| add decision tree binning encoding | T-Social · AP | 86.15 -> 98.19 | Helped | DGA-GNN: Dynamic Grouping Aggregation GNN for Fraud Detection (AAAI 2024 · 50 citations) |
| add semantic-similarity-based edges | Event2012 closed-set · ARI | 0.24 -> 0.50 | Helped | Hierarchical and Incremental Structural Entropy Minimization for Unsupervised Social Event Detection (AAAI 2024 · 46 citations) |
| Add the barely supervised learning module | Five datasets · detection performance | Helped | Barely Supervised Learning for Graph-Based Fraud Detection (AAAI 2024 · 34 citations) | |
| add Laplacian positional encodings instead of zero-vector positional features | Facebook, 10% incomplete nodes · F-Score | 0.413 -> 0.761 | Helped | GIN-SD: Source Detection in Graphs with Incomplete Nodes via Positional Encoding and Attentive Fusion (AAAI 2024 · 30 citations) |
| add RL-based neighbor selection | YelpChi · AUC | 0.9253 -> 0.9357 | Helped | DiG-In-GNN: Discriminative Feature Guided GNN-Based Fraud Detector against Inconsistencies in Multi-Relation Fraud Graph (AAAI 2024 · 26 citations) |
| add channel convolution fusion | Average (four homogeneous graphs) · AUC | 88.65 -> 89.63 | Helped | Dynamic Spectral Graph Anomaly Detection (AAAI 2025 · 25 citations) |
| Replace HALO with Euclidean distance | Amazon, Facebook, AmazonFull, YelpChiFull · AUROC | 0.7128±0.0061; 0.9251±0.0074; 0.8746±0.0026; 0.5763±0.0064 | Helped | A Label-free Heterophily-guided Approach for Unsupervised Graph Fraud Detection (AAAI 2025 · 15 citations) |
| Add category semantic decomposition (CSD) | Four fraud-detection datasets · Macro-F1 | Helped | Context-aware Graph Neural Network for Graph-based Fraud Detection with Extremely Limited Labels (AAAI 2025 · 12 citations) | |
| add cross-attention to fuse graph and text features instead of simple concatenation | UAE, Cuba, Russia, Venezuela, Iran, and China · Macro-F1 | 88.80 -> 96.43 average | Helped | IOHunter: Graph Foundation Model to Uncover Online Information Operations (AAAI 2025 · 10 citations) |
| Remove intra-level loss and replace Langevin dynamics with random feature perturbation | IBM-AML node-level · AUC | Hurt | UniFORM: Towards Unified Framework for Anomaly Detection on Graphs (AAAI 2025 · 10 citations) |
What reviewers will expect
Datasets and baselines named in the 26 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Amazon · 9 of 26 papers
- YelpChi · 9 of 26 papers
- T-Finance · 9 of 26 papers
- Reddit · 7 of 26 papers
- Elliptic · 5 of 26 papers
- Weibo · 5 of 26 papers
- Tolokers · 5 of 26 papers
- BlogCatalog · 3 of 26 papers
Baselines
- GCN · 9 of 26 papers
- DOMINANT · 8 of 26 papers
- GAT · 7 of 26 papers
- BWGNN · 7 of 26 papers
- GraphSAGE · 6 of 26 papers
- CoLA · 5 of 26 papers
- GHRN · 5 of 26 papers
- AnomalyDAE · 5 of 26 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 82 papers; from OpenAlex and the papers' affiliations).
- Monash University · 5 papers
- National University of Singapore · 5 papers
- Singapore Management University · 4 papers
- Griffith University · 4 papers
- Tianjin University · 3 papers
- Shanghai University · 3 papers
- Zhejiang University · 3 papers
- The Chinese University of Hong Kong · 3 papers
- University of Technology Sydney · 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
- Graph domain adaptation (82 papers)
- Graph out-of-distribution generalization (58 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)
- Molecular graph pretraining (101 papers)
- Brain and biological graph learning (78 papers)
- Multi-view graph clustering (70 papers)
- Graph contrastive learning analysis (61 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).