Hypergraph Benchmarks
DBLP-CA
Structure
1Graphs
41,302Nodes
22,363Edges
6Classes
1425Node Feature Dim
Coverage
21Results
1Variants
3Papers
- Heterophily-Agnostic Hypergraph Neural Networks with Riemannian Local Exchanger Heterophily-Agnostic Hypergraph Neural Networks with Riemannian Local Exchanger
- TopoBench: A Framework for Benchmarking Topological Deep Learning TopoBench: A Framework for Benchmarking Topological Deep Learning
- ICLR 2021: Hypergraph Benchmarks Hypergraph Benchmarks ICLR 2021
Node classification DBLP co-authorship hypergraph node classification (research area). Uses node-hyperedge incidence and the released node features. Clique expansion changes the input by discarding hyperedge identity. Evaluated by Accuracy.
Dataset details
Task
DBLP co-authorship hypergraph node classification (research area). The primary catalog metric is Accuracy.
Data
DBLP co-authorship hypergraph node classification (research area). The cataloged artifact reports 1 graphs.
Features
The release represents higher-order membership with a node-hyperedge incidence relation and supplies node features where available. Some source datasets have no measured features and use documented synthetic features; clique expansion is an alternative representation that discards hyperedge identity.
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
Do not silently replace incidence propagation with clique expansion or merge distinct releases sharing a short name. Random synthetic features, split seeds, hyperedge multiplicity, and isolated-node handling can dominate results and must be reported.
Sources
SOTA Progression
new SOTA
matches SOTA at that time
other proposed result
other baseline result
Leaderboard
| # | Model | Accuracy ↑ | Paper |
|---|---|---|---|
| 1 | HealHGNN | 0.9198
±
0.0031 | |
| 2 | AllSetTransformer | 0.9198
±
0.0018 | |
| 3 | ED-HNN | 0.9190
±
0.0019 | |
| 4 | PhenomNN | 0.9186
±
0.0022 | |
| 5 | Deep-HGNN | 0.9176
±
0.0028 | |
| 6 | UniGCNII | 0.9169
±
0.0019 | |
| 7 | SheafHyperGNN | 0.9159
±
0.0024 | |
| 8 | MPNN | 0.9153
±
0.0023 | |
| 9 | AllSetTransformer | 0.9153
±
0.0023 | |
| 10 | AllDeepSets | 0.9127
±
0.0027 | |
| 11 | KHGNN | 0.9121
±
0.0037 | |
| 12 | HGNN | 0.9103
±
0.0020 | |
| 13 | HCHA | 0.9092
±
0.0022 | |
| 14 | HAN (full batch) | 0.9089
±
0.0023 | |
| 15 | EDGNN | 0.9083
±
0.0025 | |
| 16 | UniGNN2 | 0.9072
±
0.0023 | |
| 17 | HyperGCN | 0.8938
±
0.0025 | |
| 18 | CEGAT | 0.8859
±
0.0029 | |
| 19 | CECGN | 0.8800
±
0.0026 | |
| 20 | HNHN | 0.8678
±
0.0029 | |
| 21 | MLP | 0.8483
±
0.0022 |
SOTA performance
set a new SOTA at the time
matched the SOTA at the time