Research map / Graph learning / Knowledge and temporal graphs
Temporal graph neural architectures
Design state-space, motif-based, and spiking architectures for dynamic graphs.
69 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Knowledge and temporal graphs, keyword cluster “dynamic · temporal graph · spatio temporal”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
24 papers in 2023–24 and 29 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 9.3% to 10.2%.
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 · Structural-RNN: Deep Learning on Spatio-Temporal Graphs (CVPR 2016 · 1,151 citations)
- 2018 · Dynamic Network Embedding by Modeling Triadic Closure Process (AAAI 2018 · 501 citations)
- 2019 · Variational Graph Recurrent Neural Networks (NeurIPS 2019 · 96 citations)
- 2020 · Inductive representation learning on temporal graphs (ICLR 2020 · 29 citations)
- 2021 · Multi-Scale Spatial Temporal Graph Convolutional Network for Skeleton-Based Action Recognition (AAAI 2021 · 335 citations)
- 2022 · AEGNN: Asynchronous Event-Based Graph Neural Networks (CVPR 2022 · 157 citations)
- 2023 · Scalable Spatiotemporal Graph Neural Networks (AAAI 2023 · 70 citations)
- 2024 · MDGNN: Multi-Relational Dynamic Graph Neural Network for Comprehensive and Dynamic Stock Investment Prediction (AAAI 2024 · 52 citations)
- 2025 · LS-TGNN: Long and Short-Term Temporal Graph Neural Network for Session-Based Recommendation (AAAI 2025 · 11 citations)
- 2026 · Trimming the Fat: Redundancy-Aware Acceleration Framework for DGNNs (AAAI 2026 · 1 citation)
Earliest papers here
- Structural-RNN: Deep Learning on Spatio-Temporal Graphs (CVPR 2016 · 1,151 citations)
- Dynamic Network Embedding by Modeling Triadic Closure Process (AAAI 2018 · 501 citations)
- Variational Graph Recurrent Neural Networks (NeurIPS 2019 · 96 citations)
What has been tried
52 papers here report 515 measured changes (a component added, removed or swapped, compared with a baseline): helped 67%, no clear effect 5%, hurt 25%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add edge weight | CSI300 · IC | 0.0268 -> 0.0322 | Helped | MDGNN: Multi-Relational Dynamic Graph Neural Network for Comprehensive and Dynamic Stock Investment Prediction (AAAI 2024 · 52 citations) |
| Add the Long-term Interest Encoder | Diginetica, Retailrocket, and Tmall · P@20 | Helped | LS-TGNN: Long and Short-Term Temporal Graph Neural Network for Session-Based Recommendation (AAAI 2025 · 11 citations) | |
| Add long-range selective modeling | clean datasets · AUC | Helped | DG-Mamba: Robust and Efficient Dynamic Graph Structure Learning with Selective State Space Models (AAAI 2025 · 8 citations) | |
| add learnable positional spike integration | Cora · accuracy | 81.93±0.54 -> 82.50±0.51 | Helped | Dynamic Reactive Spiking Graph Neural Network (AAAI 2024 · 8 citations) |
| add dropout layers | temporal centrality prediction · prediction performance | Hurt | Using Time-Aware Graph Neural Networks to Predict Temporal Centralities in Dynamic Graphs (NeurIPS 2024 · 6 citations) | |
| Remove temporal data augmentation | PEMS04 · MAE | 25.11 -> 27.31 | Hurt | ST-FiT: Inductive Spatial-Temporal Forecasting with Limited Training Data (AAAI 2025 · 5 citations) |
| add Kdhop graph building without dynamic graph building | CASIA-OHFC · SCA | 94.06 -> 94.56 | Helped | SpaceGTN: A Time-Agnostic Graph Transformer Network for Handwritten Diagram Recognition and Segmentation (AAAI 2024 · 5 citations) |
| Replace low-pass-only filtering (V-OLP) with learned 2-D spectral filters (full STSGNN) | PeMS04, PeMS08, KA-PM2.5 ablation · error (lower better) | Helped | Designing Specialized Two-Dimensional Graph Spectral Filters for Spatial-Temporal Graph Modeling (AAAI 2025 · 4 citations) | |
| Add Forman-Ricci curvature features (Spatial 3) to the spatial pattern keys | Spatio-temporal forecasting (spatial pattern analysis) · MAPE (%) | performance drop of 6.32%; MAPE of 49.97% when removed | Helped | STRAP: Spatio-Temporal Pattern Retrieval for Out-of-Distribution Generalization (NeurIPS 2025 · 4 citations) |
| Add the VCA learning objective to TGNNs | Reddit, Wikipedia, MOOC, LastFM, and GDELT · VCS | TGN: Reddit 0.18±0.02 -> 0.08±0.01; Wikipedia 0.21±0.04 -> 0.12±0.02; MOOC 0.25±0.03 -> 0.13±0.03. TCL: Reddit 0.12±0.02 | Helped | Temporal-Aware Evaluation and Learning for Temporal Graph Neural Networks (AAAI 2025 · 2 citations) |
| Replace raw text with LLM-generated text in semantic encoding | Enron transductive · MRR (%) | 49.73 ± 0.8 -> 63.07 ± 0.5; ↑13.34 | Helped | Unifying Text Semantics and Graph Structures for Temporal Text-attributed Graphs with Large Language Models (NeurIPS 2025 · 2 citations) |
| Use DZP persistence images instead of random images | Alpha (ROLAND live update) · ACC | 83.07 ± 2.44 -> *86.84 ± 1.02 | Helped | TMetaNet: Topological Meta-Learning Framework for Dynamic Link Prediction (ICML 2025 · 1 citation) |
What reviewers will expect
Datasets and baselines named in the 18 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Wikipedia · 4 of 18 papers
- Yelp · 3 of 18 papers
- Reddit · 3 of 18 papers
- PeMS03 · 2 of 18 papers
- PeMS04 · 2 of 18 papers
- PeMS08 · 2 of 18 papers
- Tmall · 2 of 18 papers
- MOOC · 2 of 18 papers
Baselines
- TGAT · 7 of 18 papers
- EvolveGCN · 4 of 18 papers
- TGN · 4 of 18 papers
- TCL · 4 of 18 papers
- GraphMixer · 4 of 18 papers
- DyRep · 3 of 18 papers
- JODIE · 3 of 18 papers
- CAWN · 3 of 18 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 53 papers; from OpenAlex and the papers' affiliations).
- Ant Group · 4 papers
- Tsinghua University · 3 papers
- City University of Hong Kong · 3 papers
- Shanghai Jiao Tong University · 3 papers
- Chinese Academy of Sciences · 2 papers
- University of Chinese Academy of Sciences · 2 papers
- Alibaba Group · 2 papers
- Sun Yat-sen University · 2 papers
- Beihang University · 2 papers
- The University of Hong Kong · 2 papers
Use it
Download BibTeX of 37 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
- Continuous-time dynamic graph learning (132 papers)
- Spatio-temporal graph traffic forecasting (60 papers)
Elsewhere in the topic
- Knowledge graph embeddings (110 papers)
- Graph reasoning for visual recognition (96 papers)
- Graph-based dialog systems (94 papers)
- Social and behavior graphs (87 papers)
- Temporal knowledge graph reasoning (84 papers)
- Scene graph generation (70 papers)
- Logical query answering and rule reasoning (68 papers)
- Heterogeneous graph learning (66 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).