Heterogeneous Graph Benchmarks
Freebase
Structure
1Graphs
180,098Nodes
1,057,688Edges
7Classes
Coverage
79Results
1Variants
10Papers
- Unified Multi-Domain Graph Pre-training for Homogeneous and Heterogeneous Graphs via Domain-Specific Expert Encoding Unified Multi-Domain Graph Pre-training for Homogeneous and Heterogeneous Graphs via Domain-Specific Expert Encoding
- ICML 2024: SlotGAT: Slot-based Message Passing for Heterogeneous Graphs SlotGAT: Slot-based Message Passing for Heterogeneous Graphs ICML 2024
- Generative-Contrastive Heterogeneous Graph Neural Network Generative-Contrastive Heterogeneous Graph Neural Network
- Graph-oriented Instruction Tuning of Large Language Models for Generic Graph Mining Graph-oriented Instruction Tuning of Large Language Models for Generic Graph Mining
- 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
- Heterogeneous Graph Contrastive Multi-view Learning Heterogeneous Graph Contrastive Multi-view Learning
- AAAI 2022: Simple and Efficient Heterogeneous Graph Neural Network Simple and Efficient Heterogeneous Graph Neural Network AAAI 2022
- KDD 2021: Heterogeneous Graph Benchmarks Heterogeneous Graph Benchmarks KDD 2021
- Leveraging Meta-path Contexts for Classification in Heterogeneous Information Networks Leveraging Meta-path Contexts for Classification in Heterogeneous Information Networks
Node classification Heterogeneous Freebase knowledge graph node classification (8 entity types). Uses typed nodes, typed relations, and available type-specific features. Evaluated by Macro-F1.
Dataset details
Task
Heterogeneous Freebase knowledge graph node classification (8 entity types). The primary catalog metric is Macro-F1.
Data
Heterogeneous Freebase knowledge graph node classification (8 entity types).
Features
Inputs are typed graphs with relation-specific edges and type-specific feature matrices where available. Some node types lack native features, so learned embeddings or propagated features are modeling choices rather than measured attributes.
Splits and evaluation
The catalog records these protocols or variants: Standard split. Evaluation uses Macro-F1 (higher is better). Exact masks or folds must come from the cited release.
Quirks and pitfalls
Schema, reverse-edge creation, feature propagation, and negative sampling materially affect results. Report the exact release and target type; random edge splits may leak future or reciprocal information.
Sources
SOTA Progression
new SOTA
matches SOTA at that time
other proposed result
other baseline result
Leaderboard
| # | Model | Macro-F1 ↑ | Micro-F1 | Paper |
|---|---|---|---|---|
| 1 | ConCH | 0.6475 | — | |
| 2 | HetGNN | 0.6292 | — | |
| 3 | GC-HGNN | 0.6247
±
0.0042 | — | |
| 4 | MAGNN | 0.6223 | — | |
| 5 | HGT | 0.6176 | — | |
| 6 | HDGI | 0.6139 | — | |
| 7 | GCN | 0.6090 | — | |
| 8 | HGMAE | 0.6082
±
0.0051 | — | |
| 9 | HGCN | 0.6035 | — | |
| 10 | HAN | 0.6011 | — | |
| 11 | HeCo | 0.5987
±
0.0033 | — | |
| 12 | MuseGraph (LLaMA3-8B) | 0.5962 | — | |
| 13 | HERec | 0.5946
±
0.0116 | — | |
| 14 | DeepWalk | 0.5920
±
0.0050 | — | |
| 15 | HGCML | 0.5892
±
0.0257 | — | |
| 16 | HAN | 0.5850
±
0.0093 | — | |
| 17 | HGT | 0.5830
±
0.0072 | — | |
| 18 | Mp2vec | 0.5750 | — | |
| 19 | Node2Vec | 0.5735 | — | |
| 20 | MHGCN | 0.5712
±
0.0032 | — | |
| 21 | DMGI | 0.5599
±
0.0059 | — | |
| 22 | Mp2vec | 0.5550
±
0.0049 | — | |
| 23 | RpHGNN | 0.5402
±
0.0088 | — | |
| 24 | LMSPS | 0.5326
±
0.0047 | — | |
| 25 | HINormer | 0.5218
±
0.0039 | — | |
| 26 | SeHGNN | 0.5187
±
0.0086 | — | |
| 27 | MVGRL | 0.5115 | — | |
| 28 | GAT | 0.5109 | — | |
| 29 | NARS | 0.4998
±
0.0177 | — | |
| 30 | SlotGAT | 0.4968
±
0.0197 | — | |
| 31 | HINormer | 0.4887 | — | |
| 32 | HGB | 0.4772
±
0.0148 | — | |
| 33 | Simple-HGN | 0.4772
±
0.0148 | — | |
| 34 | simpleHGN | 0.4772
±
0.0148 | — | |
| 35 | R-GCN | 0.4703 | — | |
| 36 | RGCN | 0.4678
±
0.0077 | — | |
| 37 | GPH^2 | 0.4478
±
0.0547 | — | |
| 38 | GraphSAGE | 0.4417
±
0.0114 | — | |
| 39 | space4HGNN | 0.4137
±
0.0449 | — | |
| 40 | HGPrompt | 0.4078
±
0.0323 | — | |
| 41 | GAT | 0.4074
±
0.0258 | — | |
| 42 | HeCo | 0.3799
±
0.0312 | — | |
| 43 | HGMAE | 0.3515
±
0.0362 | — | |
| 44 | OFA | 0.3475 | — | |
| 45 | DMGI | 0.3337
±
0.0232 | — | |
| 46 | HERO | 0.3139
±
0.0780 | — | |
| 47 | HGT | 0.2928
±
0.0252 | — | |
| 48 | HetGPT | 0.2903
±
0.0654 | — | |
| 49 | GCN | 0.2784
±
0.0313 | — | |
| 50 | GraphSAGE | 0.2759 | — |
SOTA performance
set a new SOTA at the time
matched the SOTA at the time