LINKX Benchmarks
Pokec
Structure
1Graphs
1,632,803Nodes
30,622,564Edges
2Classes
65Node Feature Dim
0.445Homophily
Coverage
70Results
1Variants
9Papers
- Robust Learning on Heterogeneous Graphs with Heterophily: A Graph Structure Learning Approach Robust Learning on Heterogeneous Graphs with Heterophily: A Graph Structure Learning Approach
- Graph Learning at Scale: Characterizing and Optimizing Pre-Propagation GNNs Graph Learning at Scale: Characterizing and Optimizing Pre-Propagation GNNs
- ICLR 2024: Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning ICLR 2024
- ICLR 2023: Mixture of Weak & Strong Experts on Graphs Mixture of Weak & Strong Experts on Graphs ICLR 2023
- ICML 2023: Graph Neural Networks with Learnable and Optimal Polynomial Bases Graph Neural Networks with Learnable and Optimal Polynomial Bases ICML 2023
- ICLR 2023: Discovering Data Geometry: Energy-Driven Neural Diffusion Discovering Data Geometry: Energy-Driven Neural Diffusion ICLR 2023
- AAAI 2022: Interpreting Unfairness in Graph Neural Networks via Training Node Attribution Interpreting Unfairness in Graph Neural Networks via Training Node Attribution AAAI 2022
- FMP: Toward Fair Graph Message Passing against Topology Bias FMP: Toward Fair Graph Message Passing against Topology Bias
- NeurIPS 2022: Convolutional Neural Networks on Graphs with Chebyshev Approximation, Revisited Convolutional Neural Networks on Graphs with Chebyshev Approximation, Revisited NeurIPS 2022
Binary node classification Slovak social network Pokec node classification (binary gender, 1.6M nodes). Uses the released sparse node features and a low-homophily graph. Feature-only baselines are essential because structure can be weakly predictive. Evaluated by Accuracy.
Dataset details
Task
Slovak social network Pokec node classification (binary gender, 1.6M nodes). The primary catalog metric is Accuracy.
Data
Slovak social network Pokec node classification (binary gender, 1.6M nodes). The cataloged artifact reports 1 graphs, 2 target classes or tasks.
Features
Each release supplies a sparse node-feature matrix, graph edges, and node labels. Feature meaning varies by source; structure and features are deliberately evaluated without assuming neighboring labels agree.
Splits and evaluation
The catalog records these protocols or variants: Standard split. Evaluation uses Accuracy (higher is better). Exact masks or folds must come from the cited release.
Quirks and pitfalls
These are transductive single-graph tasks with fixed random splits. Label homophily is low and class balance varies; feature-only and structure-only baselines are essential, and similarly named OGB/source datasets may use different labels or splits.
Sources
SOTA Progression
new SOTA
matches SOTA at that time
other proposed result
other baseline result
Leaderboard
| # | Model | Accuracy ↑ | Paper |
|---|---|---|---|
| 1 | SAGE | 0.8597
±
0.0021 | |
| 2 | Polynormer | 0.8595
±
0.0007 | |
| 3 | NID_SAGE | 0.8563
±
0.0031 | |
| 4 | MoE-H2GCN | 0.8302
±
0.0030 | |
| 5 | OptBasisGNN | 0.8283
±
0.0004 | |
| 6 | ChebNetII | 0.8233
±
0.0028 | |
| 7 | LINKX | 0.8204
±
0.0007 | |
| 8 | HOGA | 0.8191 | |
| 9 | BernNet | 0.8167
±
0.0017 | |
| 10 | MixHop | 0.8107
±
0.0016 | |
| 11 | SIGN | 0.8101 | |
| 12 | H2GCN | 0.8089
±
0.0016 | |
| 13 | GPR-GNN | 0.8074
±
0.0022 | |
| 14 | LINK | 0.8054
±
0.0003 | |
| 15 | MoE-GIN-skip | 0.7970
±
0.0023 | |
| 16 | GCNII | 0.7894
±
0.0011 | |
| 17 | GPR-GNN | 0.7883
±
0.0005 | |
| 18 | MoE-SAGE | 0.7784
±
0.0004 | |
| 19 | MoE-GCN | 0.7728
±
0.0008 | |
| 20 | GraphMoE-GCN | 0.7699
±
0.0010 | |
| 21 | GraphMoE-SAGE | 0.7667
±
0.0008 | |
| 22 | GIN-skip | 0.7629
±
0.0051 | |
| 23 | MoE-GIN | 0.7611
±
0.0039 | |
| 24 | GCN | 0.7601
±
0.0049 | |
| 25 | GraphSAGE | 0.7576
±
0.0004 | |
| 26 | GCN | 0.7545
±
0.0017 | |
| 27 | SGFormer | 0.7376
±
0.0024 | |
| 28 | GCN-skip | 0.7346
±
0.0004 | |
| 29 | ChebNet | 0.7271
±
0.0066 | |
| 30 | GCN | 0.7178
±
0.0037 | |
| 31 | SGC | 0.7124
±
0.0046 | |
| 32 | FMP | 0.7050
±
0.0050 | |
| 33 | MLP | 0.7048
±
0.0077 | |
| 34 | GAT | 0.6976
±
0.0130 | |
| 35 | DIFFormer-s | 0.6924
±
0.0076 | |
| 36 | NodeFormer | 0.6832
±
0.0045 | |
| 37 | APPNP | 0.6691
±
0.0146 | |
| 38 | GAT | 0.6557
±
0.0034 | |
| 39 | NIFTY | 0.6530
±
0.0020 | |
| 40 | FairGNN | 0.6400
±
0.0070 | |
| 41 | BIND 1% | 0.6350
±
0.0040 | |
| 42 | Van. GCN | 0.6320
±
0.0070 | |
| 43 | BIND 10% | 0.6290
±
0.0040 | |
| 44 | MLP | 0.6237
±
0.0002 | |
| 45 | GCN | 0.6231
±
0.0113 | |
| 46 | EDITS | 0.6160
±
0.0090 | |
| 47 | MLP | 0.6015
±
0.0003 | |
| 48 | AdaGCN | 0.5592
±
0.0035 | |
| 49 | GIN | 0.5337
±
0.0215 | |
| 50 | LP | 0.5273 |
SOTA performance
set a new SOTA at the time
matched the SOTA at the time