OGB
ogbn-mag
Structure
1Graphs
1,939,743Nodes
21,111,007Edges
349Classes
128Node Feature Dim
0.9205Heterophily
Coverage
106Results
1Variants
17Papers
- Robust Learning on Heterogeneous Graphs with Heterophily: A Graph Structure Learning Approach Robust Learning on Heterogeneous Graphs with Heterophily: A Graph Structure Learning Approach
- TMLR 2025: Towards Efficient Training of Graph Neural Networks: A Multiscale Approach Towards Efficient Training of Graph Neural Networks: A Multiscale Approach TMLR 2025
- TMLR 2024: SCNode: Spatial and Contextual Coordinates for Graph Representation Learning SCNode: Spatial and Contextual Coordinates for Graph Representation Learning TMLR 2024
- Rethinking and Simplifying Bootstrapped Graph Latents Rethinking and Simplifying Bootstrapped Graph Latents
- Efficient Heterogeneous Graph Learning via Random Projection Efficient Heterogeneous Graph Learning via Random Projection
- NeurIPS 2023: Long-range Meta-path Search on Large-scale Heterogeneous Graphs Long-range Meta-path Search on Large-scale Heterogeneous Graphs NeurIPS 2023
- KDD 2023: Improving Expressivity of GNNs with Subgraph-specific Factor Embedded Normalization Improving Expressivity of GNNs with Subgraph-specific Factor Embedded Normalization KDD 2023
- AAAI 2022: Simple and Efficient Heterogeneous Graph Neural Network Simple and Efficient Heterogeneous Graph Neural Network AAAI 2022
- Label-Enhanced Graph Neural Network for Semi-supervised Node Classification Label-Enhanced Graph Neural Network for Semi-supervised Node Classification
- SCR: Training Graph Neural Networks with Consistency Regularization SCR: Training Graph Neural Networks with Consistency Regularization
- Adaptive Label Smoothing To Regularize Large-Scale Graph Training Adaptive Label Smoothing To Regularize Large-Scale Graph Training
- Graph Attention MLP with Reliable Label Utilization Graph Attention MLP with Reliable Label Utilization
- Probing GNN Explainers: A Rigorous Theoretical and Empirical Analysis of GNN Explanation Methods Probing GNN Explainers: A Rigorous Theoretical and Empirical Analysis of GNN Explanation Methods
- Heterogeneous Graph Representation Learning with Relation Awareness Heterogeneous Graph Representation Learning with Relation Awareness
- Residual Network and Embedding Usage: New Tricks of Node Classification with Graph Convolutional Networks Residual Network and Embedding Usage: New Tricks of Node Classification with Graph Convolutional Networks
- Scalable and Adaptive Graph Neural Networks with Self-Label-Enhanced Training Scalable and Adaptive Graph Neural Networks with Self-Label-Enhanced Training
- Scalable Graph Neural Networks for Heterogeneous Graphs Scalable Graph Neural Networks for Heterogeneous Graphs
349-class node classification Predict the conference or journal of each paper using paper content, citations, authors, affiliations, and fields of study. Evaluated by Accuracy.
Dataset details
Task
Predict the conference or journal of each paper using paper content, citations, authors, affiliations, and fields of study. It is a 349-class paper-node classification task scored by accuracy.
Data
The heterogeneous graph has paper, author, institution, and field-of-study nodes, connected by writes, affiliation, citation, and topic relations. It contains 736,389 papers within 1,939,743 total entities.
Features
Only paper nodes receive standard input features, each a 128-dimensional average word2vec representation of title and abstract. Other entity types must obtain representations from identifiers or graph structure rather than supplied content features.
Splits and evaluation
Papers published before 2018 train the model, 2018 papers validate it, and papers from 2019 onward form test. The complete heterogeneous topology is present in the transductive graph.
Quirks and pitfalls
Venue labels in MAG can be noisy or missing, and the graph snapshot is no longer a live reflection of the scholarly record. Collapsing relation or node types changes the benchmark. Methods using additional author, venue, or text metadata must distinguish it from the standard inputs.
Sources
SOTA Progression
new SOTA
matches SOTA at that time
other proposed result
other baseline result
Leaderboard
| # | Model | Accuracy ↑ | Instability | Unfaithfulness | Paper |
|---|---|---|---|---|---|
| 1 | LMSPS+LP+MS | 0.5784
±
0.0022 | — | — | |
| 2 | SeHGNN+emb+ms | 0.5719
±
0.0012 | — | — | |
| 3 | SeHGNN+ms | 0.5671
±
0.0014 | — | — | |
| 4 | SeHGNN+LP+MS | 0.5671
±
0.0014 | — | — | |
| 5 | RLU+SCR | 0.5631
±
0.0021 | — | — | |
| 6 | RLU | 0.5590
±
0.0027 | — | — | |
| 7 | GAMLP+emb+ms | 0.5590
±
0.0027 | — | — | |
| 8 | NARS-GAMLP+RLU | 0.5590
±
0.0027 | — | — | |
| 9 | LMSPS+LP | 0.5510
±
0.0011 | — | — | |
| 10 | LMSPS | 0.5483
±
0.0020 | — | — | |
| 11 | SeHGNN+emb | 0.5478
±
0.0017 | — | — | |
| 12 | SCR-m | 0.5451
±
0.0019 | — | — | |
| 13 | SLE | 0.5440
±
0.0015 | — | — | |
| 14 | SAGN+emb+ms | 0.5440
±
0.0015 | — | — | |
| 15 | NARS-SAGN+2-SLE | 0.5440
±
0.0015 | — | — | |
| 16 | SCR | 0.5432
±
0.0018 | — | — | |
| 17 | HGUL | 0.5421
±
0.0027 | — | — | |
| 18 | SeHGNN | 0.5399
±
0.0018 | — | — | |
| 19 | SeHGNN+LP | 0.5399
±
0.0018 | — | — | |
| 20 | NARS_GAMLP | 0.5396
±
0.0018 | — | — | |
| 21 | GAMLP+emb | 0.5396
±
0.0018 | — | — | |
| 22 | NARS-GAMLP | 0.5396
±
0.0018 | — | — | |
| 23 | LEGNN + AS-Train | 0.5378
±
0.0016 | — | — | |
| 24 | LEGNN | 0.5378 | — | — | |
| 25 | LEGNN with SuperNorm | 0.5324
±
0.0013 | — | — | |
| 26 | H2G-former | 0.5312
±
0.0032 | — | — | |
| 27 | SIGN | 0.5289
±
0.0011 | — | — | |
| 28 | LEGNN | 0.5276
±
0.0014 | — | — | |
| 29 | NARS+emb | 0.5240
±
0.0016 | — | — | |
| 30 | NARS | 0.5240
±
0.0016 | — | — | |
| 31 | NARS_SAGN | 0.5232
±
0.0025 | — | — | |
| 32 | NARS-SAGN+0-SLE | 0.5232
±
0.0025 | — | — | |
| 33 | RpHGNN | 0.5207
±
0.0017 | — | — | |
| 34 | R-HGNN | 0.5204 | — | — | |
| 35 | GAMLP | 0.5163
±
0.0022 | — | — | |
| 36 | GAMLP+LP | 0.5163
±
0.0022 | — | — | |
| 37 | Hetero2Net | 0.5160
±
0.0033 | — | — | |
| 38 | SeHGNN | 0.5145
±
0.0029 | — | — | |
| 39 | SAGN | 0.5117
±
0.0032 | — | — | |
| 40 | SAGN+LP | 0.5117
±
0.0032 | — | — | |
| 41 | R-GraphSAGE | 0.5094
±
0.0044 | — | — | |
| 42 | NARS | 0.5088
±
0.0012 | — | — | |
| 43 | NARS | 0.5066
±
0.0022 | — | — | |
| 44 | HGConv | 0.5045
±
0.0017 | — | — | |
| 45 | R-GSN | 0.5032
±
0.0037 | — | — | |
| 46 | HGT | 0.5023
±
0.0048 | — | — | |
| 47 | SCNode | 0.5003 | — | — | |
| 48 | HGT+emb | 0.4982
±
0.0013 | — | — | |
| 49 | GraphSAINT + metapath2vec | 0.4969
±
0.0022 | — | — | |
| 50 | HGT | 0.4921 | — | — |
SOTA performance
set a new SOTA at the time
matched the SOTA at the time