OGB
ogbn-papers100M
Structure
1Graphs
111,059,956Nodes
1,615,685,872Edges
172Classes
128Node Feature Dim
Coverage
130Results
1Variants
32Papers
- Coden: Efficient Temporal Graph Neural Networks for Continuous Prediction Coden: Efficient Temporal Graph Neural Networks for Continuous Prediction
- KDD 2025: Partition-wise Graph Filtering: A Unified Perspective Through the Lens of Graph Coarsening Partition-wise Graph Filtering: A Unified Perspective Through the Lens of Graph Coarsening KDD 2025
- Graph Learning at Scale: Characterizing and Optimizing Pre-Propagation GNNs Graph Learning at Scale: Characterizing and Optimizing Pre-Propagation GNNs
- ICLR 2025: Accurate and Scalable Graph Neural Networks via Message Invariance Accurate and Scalable Graph Neural Networks via Message Invariance ICLR 2025
- WWW 2024: Polynomial Selection in Spectral Graph Neural Networks: An Error-Sum of Function Slices Approach Polynomial Selection in Spectral Graph Neural Networks: An Error-Sum of Function Slices Approach WWW 2024
- ICML 2024: Unifews: You Need Fewer Operations for Efficient Graph Neural Networks Unifews: You Need Fewer Operations for Efficient Graph Neural Networks ICML 2024
- KDD 2024: UniGraph: Learning a Unified Cross-Domain Foundation Model for Text-Attributed Graphs UniGraph: Learning a Unified Cross-Domain Foundation Model for Text-Attributed Graphs KDD 2024
- WWW 2024: Rethinking Node-wise Propagation for Large-scale Graph Learning Rethinking Node-wise Propagation for Large-scale Graph Learning WWW 2024
- LightDiC: A Simple yet Effective Approach for Large-scale Digraph Representation Learning LightDiC: A Simple yet Effective Approach for Large-scale Digraph Representation Learning
- WWW 2024: Disentangled Condensation for Large-scale Graphs Disentangled Condensation for Large-scale Graphs WWW 2024
- Graph Transformers for Large Graphs Graph Transformers for Large Graphs
- ICLR 2023: MuseGNN: Forming Scalable, Convergent GNN Layers that Minimize a Sampling-Based Energy MuseGNN: Forming Scalable, Convergent GNN Layers that Minimize a Sampling-Based Energy ICLR 2023
- FedGTA: Topology-aware Averaging for Federated Graph Learning FedGTA: Topology-aware Averaging for Federated Graph Learning
- WWW 2023: Node-wise Diffusion for Scalable Graph Learning Node-wise Diffusion for Scalable Graph Learning WWW 2023
- ICML 2023: Graph Neural Networks with Learnable and Optimal Polynomial Bases Graph Neural Networks with Learnable and Optimal Polynomial Bases ICML 2023
- ICLR 2022: Learning on Large-scale Text-attributed Graphs via Variational Inference Learning on Large-scale Text-attributed Graphs via Variational Inference ICLR 2022
- SCARA: Scalable Graph Neural Networks with Feature-Oriented Optimization SCARA: Scalable Graph Neural Networks with Feature-Oriented Optimization
- KDD 2022: Graph Attention Multi-Layer Perceptron Graph Attention Multi-Layer Perceptron KDD 2022
- KDD 2022: Model Degradation Hinders Deep Graph Neural Networks Model Degradation Hinders Deep Graph Neural Networks KDD 2022
- NeurIPS 2022: Rethinking and Scaling Up Graph Contrastive Learning: An Extremely Efficient Approach with Group Discrimination Rethinking and Scaling Up Graph Contrastive Learning: An Extremely Efficient Approach with Group Discrimination NeurIPS 2022
- Neighbor2Seq: Deep Learning on Massive Graphs by Transforming Neighbors to Sequences Neighbor2Seq: Deep Learning on Massive Graphs by Transforming Neighbors to Sequences
- NeurIPS 2022: Convolutional Neural Networks on Graphs with Chebyshev Approximation, Revisited Convolutional Neural Networks on Graphs with Chebyshev Approximation, Revisited NeurIPS 2022
- SCR: Training Graph Neural Networks with Consistency Regularization SCR: Training Graph Neural Networks with Consistency Regularization
- Sequential Aggregation and Rematerialization: Distributed Full-batch Training of Graph Neural Networks on Large Graphs Sequential Aggregation and Rematerialization: Distributed Full-batch Training of Graph Neural Networks on Large Graphs
- ICLR 2021: Node Feature Extraction by Self-Supervised Multi-scale Neighborhood Prediction Node Feature Extraction by Self-Supervised Multi-scale Neighborhood Prediction ICLR 2021
- NeurIPS 2021: Node Dependent Local Smoothing for Scalable Graph Learning Node Dependent Local Smoothing for Scalable Graph Learning NeurIPS 2021
- Graph Attention MLP with Reliable Label Utilization Graph Attention MLP with Reliable Label Utilization
- Evaluating Deep Graph Neural Networks Evaluating Deep Graph Neural Networks
- Grain: Improving Data Efficiency of Graph Neural Networks via Diversified Influence Maximization Grain: Improving Data Efficiency of Graph Neural Networks via Diversified Influence Maximization
- Scalable and Adaptive Graph Neural Networks with Self-Label-Enhanced Training Scalable and Adaptive Graph Neural Networks with Self-Label-Enhanced Training
- DistGNN: Scalable Distributed Training for Large-Scale Graph Neural Networks DistGNN: Scalable Distributed Training for Large-Scale Graph Neural Networks
- SIGN: Scalable Inception Graph Neural Networks SIGN: Scalable Inception Graph Neural Networks
172-class node classification Predict a primary arXiv subject area for the approximately 1.5 million labeled arXiv papers while using the full 111-million-node citation network. Evaluated by Accuracy.
Dataset details
Task
Predict a primary arXiv subject area for the approximately 1.5 million labeled arXiv papers while using the full 111-million-node citation network. Accuracy over 172 classes is the official measure.
Data
Nodes are papers indexed by Microsoft Academic Graph and directed edges are citations. Most nodes are not arXiv papers and have no class label, but remain available as unlabeled graph context.
Features
As in ogbn-arxiv, every paper has a 128-dimensional average of skip-gram embeddings for title and abstract words. The fixed features and graph come from the released MAG snapshot.
Splits and evaluation
Labeled arXiv papers through 2017 train the model, papers from 2018 validate it, and papers from 2019 onward form test. The task is transductive because features and topology for all nodes are supplied.
Quirks and pitfalls
Dataset scale makes full-batch processing impractical and often forces sampling or partitioning choices that affect results. The overwhelming majority of nodes are unlabeled. This is not simply a scaled ogbn-arxiv: it uses 172 subjects and a much broader MAG graph.
Sources
SOTA Progression
new SOTA
matches SOTA at that time
other proposed result
other baseline result
Leaderboard
| # | Model | Accuracy ↑ | F1 | Paper |
|---|---|---|---|---|
| 1 | SGC | 0.7910 | — | |
| 2 | GAMLP | 0.7036
±
0.0002 | — | |
| 3 | GLEM-GNN | 0.7036
±
0.0002 | — | |
| 4 | ShaDow+ATP | 0.6980
±
0.0010 | — | |
| 5 | GAMLP+RLU | 0.6967
±
0.0004 | — | |
| 6 | CPF | 0.6887
±
0.0030 | — | |
| 7 | TFGNN | 0.6865
±
0.0020 | — | |
| 8 | RLU+SCR | 0.6842
±
0.0015 | — | |
| 9 | SLE | 0.6830
±
0.0008 | — | |
| 10 | RLU | 0.6825
±
0.0011 | — | |
| 11 | GAMLP+RLU | 0.6825
±
0.0011 | — | |
| 12 | SCR-m | 0.6816
±
0.0012 | — | |
| 13 | SCR | 0.6814
±
0.0008 | — | |
| 14 | GAMLP(JK) | 0.6807
±
0.0010 | — | |
| 15 | SAGN+SLE | 0.6800
±
0.0015 | — | |
| 16 | UniGraph | 0.6789
±
0.0021 | — | |
| 17 | GAMLP | 0.6771
±
0.0020 | — | |
| 18 | SAGN+SLE | 0.6755
±
0.0015 | — | |
| 19 | UniMP | 0.6736
±
0.0010 | — | |
| 20 | SGC+AIR | 0.6723
±
0.0020 | — | |
| 21 | OptBasisGNN | 0.6722
±
0.0015 | — | |
| 22 | TOP-GCNII | 0.6721
±
0.0012 | — | |
| 23 | GraphSAINT+ATP | 0.6720
±
0.0020 | — | |
| 24 | ChebNetII | 0.6718
±
0.0032 | — | |
| 25 | shaDow | 0.6710
±
0.0020 | — | |
| 26 | AdaptKry | 0.6707
±
0.0020 | — | |
| 27 | SIGN-XL | 0.6706
±
0.0017 | — | |
| 28 | OptBasis | 0.6703
±
0.0030 | — | |
| 29 | HOGA | 0.6686
±
0.0011 | — | |
| 30 | MuseGNN | 0.6682 | — | |
| 31 | UniFilter | 0.6679
±
0.0030 | — | |
| 32 | SAGN | 0.6675
±
0.0084 | — | |
| 33 | GLEM-LM | 0.6671
±
0.0025 | — | |
| 34 | NIGCN | 0.6657
±
0.0020 | — | |
| 35 | Supervised SGC | 0.6650
±
0.0020 | — | |
| 36 | GAMLP (FedGTA) | 0.6650
±
0.0030 | — | |
| 37 | Coden | 0.6645
±
0.0033 | — | |
| 38 | SAGE | 0.6644
±
0.0017 | — | |
| 39 | Cluster-GCN+ATP | 0.6640
±
0.0020 | — | |
| 40 | NFGNN | 0.6638
±
0.0020 | — | |
| 41 | SIGN | 0.6636
±
0.0005 | — | |
| 42 | GraphSage+C&S | 0.6630 | — | |
| 43 | GAT (NS) | 0.6628
±
0.0008 | — | |
| 44 | GraphMAE2 | 0.6624
±
0.0054 | — | |
| 45 | FreshGNN | 0.6622
±
0.0007 | — | |
| 46 | SAGE (NS) | 0.6620
±
0.0013 | — | |
| 47 | PCConv | 0.6619
±
0.0020 | — | |
| 48 | GPRGNN | 0.6613
±
0.0020 | — | |
| 49 | Node-MoE | 0.6612
±
0.0020 | — | |
| 50 | SGC | 0.6610
±
0.0013 | — |
SOTA performance
set a new SOTA at the time
matched the SOTA at the time