启智社区（中文版） | OpenHGNN [CIKM2022] | Space4HGNN [SIGIR2022] | Benchmark&Leaderboard | Slack Channel
This is an open-source toolkit for Heterogeneous Graph Neural Network based on DGL [Deep Graph Library] and PyTorch. We integrate SOTA models of heterogeneous graph.
Latest version: OpenHGNN v0.9.0. v0.9 adds 10 model contributions, the node_regression task, model reproduction documentation, and clearer DGL-based model integration guidance.
2026-06-26 release v0.9
We release OpenHGNN v0.9.
- Added 10 model contributions, corresponding to 11 registered model names: HGDL, HGEN, HGSketch, HGOT, RMR, HERO/HERO_homo, SEHTGNN, HTGformer, HCMGNN, and RelGT.
- The current version registers 83 model names and 17 task/flow entries.
- Added the
node_regressiontask for continuous node-label prediction. - Updated documentation entry points, quick start, model overview, task overview, and model PR checklist.
- Strengthened model contribution standards around DGL implementation, trainerflow, dataset, README, and smoke-test requirements.
2024-07-23 release v0.7
We released OpenHGNN v0.7.0.
- New models and datasets.
- Graph Prompt pipeline
- Data process frame: dgl.graphBolt
- New GNN aggregator: dgl.sparse
- Distributed training
2023-07-17 release v0.5
We released OpenHGNN v0.5.0.
- New models and datasets.
- 4 New tasks: pretrain, recommendation, graph attacks and defenses, abnorm_event detection.
- TensorBoard visualization.
- Maintenance and test module.
2023-02-24 OpenI Excellent Incubation Award
OpenHGNN won the Excellent Incubation Program Award of OpenI Community! For more details：https://mp.weixin.qq.com/s/PpbwEdP0-8wG9dsvRvRDaA
2023-02-21 First Prize of CIE
The algorithm library supports the project of "Intelligent Analysis Technology and Scale Application of Large Scale Complex Heterogeneous Graph Data" led by BUPT and participated by ANT GROUP, China Mobile, Haizhi Technology, etc. This project won the first prize of the 2022 Chinese Intitute of Electronics "Science and Technology Progress Award".
2023-01-13 release v0.4
We released OpenHGNN v0.4.
- New models
- Provide pipelines for applications
- More models supporting mini-batch training
- Benchmark for million-scale graphs
2022-08-02 paper accepted
Our paper [ OpenHGNN: An Open Source Toolkit for Heterogeneous Graph Neural Network ](https://dl.acm.org/doi/abs/10.1145/3511808.3557664) is accpeted at CIKM 2022 short paper track.
2022-06-27 release v0.3
We released OpenHGNN v0.3.
- New models
- API Usage
- Simply customization of user-defined datasets and models
- Visualization tools of heterogeneous graphs
2022-02-28 release v0.2
We released OpenHGNN v0.2.
- New Models
- Space4HGNN [SIGIR2022]
- Benchmark&Leaderboard
2022-01-07 加入启智社区
启智社区用户可以享受到如下功能：
- 全新的中文文档
- 免费的计算资源—— 云脑使用教程
- OpenHGNN最新功能
- 新增模型：【KDD2017】Metapath2vec、【TKDE2018】HERec、【KDD2021】HeCo、【KDD2021】SimpleHGN、【TKDE2021】HPN、【ICDM2021】HDE、fastGTN
- 新增日志功能
- 新增美团外卖数据集
- Easy-to-Use: OpenHGNN provides easy-to-use interfaces for running experiments with the given models and dataset. Besides, we also integrate optuna to get hyperparameter optimization.
- Extensibility: User can define customized task/model/dataset to apply new models to new scenarios.
- Efficiency: The backend dgl provides efficient APIs.
- Discoverability: CLI commands can inspect the environment, registered models, registered tasks, and datasets.
Current version statistics:
- Registered model names: 83.
- Registered task/flow entries: 17.
- v0.9 model contributions: 10.
HEROandHERO_homoare two registered names for one model contribution. - v0.9 task addition:
node_regression.
Common documentation entry points:
- v0.9 release notes
- Quick start
- Model overview
- Model reproduction guide
- Task overview and node_regression
- Model PR checklist
1. Python environment (Optional): We recommend using Conda package manager
Officially recommended environments:
- Primary:
Python 3.11 + PyTorch 2.4.0 + DGL 2.4.0+cu121 - Compatibility:
Python 3.10 + PyTorch 2.3.1 + DGL 2.2.1
The repository file environment.yml is pinned to the primary setup.
conda create -n openhgnn python=3.11
conda activate openhgnn2. Install Pytorch: Follow their tutorial to run the proper command according to your OS and CUDA version. For example:
pip install torch==2.4.0 torchvision torchaudio3. Install DGL: Follow their tutorial to run the proper command according to your OS and CUDA version. For example:
pip install dgl==2.4.0+cu121 -f https://data.dgl.ai/wheels/torch-2.4/cu121/repo.html4. Install openhgnn:
- install from pypi
pip install openhgnn- install from source
git clone https://github.com/BUPT-GAMMA/OpenHGNN
# If you encounter a network error, try git clone from openi as following.
# git clone https://git.openi.org.cn/GAMMALab/OpenHGNN.git
cd OpenHGNN
# To create the full pinned source environment directly, you can also use:
# conda env create -f environment.yml && conda activate openhgnn
pip install -r requirements.txt
pip install -e .You can also use the packaged CLI to inspect the supported registry and current environment:
openhgnn list models
openhgnn list tasks
openhgnn env --format json5. Install gdbi(Optional):
- install gdbi from git
pip install git+https://github.com/xy-Ji/gdbi.git- install graph database from pypi
pip install neo4j==5.16.0
pip install nebula3-python==3.4.0Running an existing baseline model on an existing benchmark dataset
python main.py -m model_name -d dataset_name -t task_name -g 0 --use_best_config --load_from_pretrainedusage: main.py [-h] [--model MODEL] [--task TASK] [--dataset DATASET] [--gpu GPU] [--use_best_config][--use_database]
optional arguments:
-h, --help show this help message and exit
--model -m name of models
--task -t name of task
--dataset -d name of datasets
--gpu -g controls which gpu you will use. If you do not have gpu, set -g -1.
--use_best_config use_best_config means you can use the best config in the dataset with the model. If you want to
set the different hyper-parameter, modify the openhgnn.config.ini manually. The best_config
will override the parameter in config.ini.
--load_from_pretrained will load the model from a default checkpoint.
--use_database get dataset from database
---mini_batch_flag train model with mini-batchs
---graphbolt mini-batch training with dgl.graphbolt
---use_distributed train model with distributed way
e.g.:
python main.py -m GTN -d imdb4GTN -t node_classification -g 0 --use_best_config
python main.py -m RGCN -d imdb4GTN -t node_classification -g 0 --mini_batch_flag --graphboltNote: If you are interested in some model, you can refer to the below models list.
Refer to the docs to get more basic and depth usage.
tensorboard --logdir=./openhgnn/output/{model_name}/e.g.：
tensorboard --logdir=./openhgnn/output/RGCN/Note: To visualize results, you need to train the model first.
take neo4j and imdb dataset for example
- construct csv file for dataset(node-level:A.csv,edge-level:A_P.csv)
- import csv file to database
LOAD CSV WITH HEADERS FROM "file:///data.csv" AS row
CREATE (:graphname_labelname {ID: row.ID, ... });- add user information to access database in config.py file
self.graph_address = [graph_address]
self.user_name = [user_name]
self.password = [password]- e.g.:
python main.py -m MAGNN -d imdb4MAGNN -t node_classification -g 0 --use_best_config --use_databaseThe current version registers 83 model names. The table below tracks the main model-task coverage maintained in README. Use openhgnn list models for the complete registered model list.
| Model | Node classification | Link prediction | Recommendation | Node regression | Notes |
|---|---|---|---|---|---|
| TransE[NIPS 2013] | ✔️ | Knowledge graph embedding | |||
| TransH[AAAI 2014] | ✔️ | Knowledge graph embedding | |||
| TransR[AAAI 2015] | ✔️ | Knowledge graph embedding | |||
| TransD[ACL 2015] | ✔️ | Knowledge graph embedding | |||
| Metapath2vec[KDD 2017] | ✔️ | Representation learning | |||
| RGCN[ESWC 2018] | ✔️ | ✔️ | |||
| HERec[TKDE 2018] | ✔️ | Representation learning | |||
| HAN[WWW 2019] | ✔️ | ✔️ | |||
| KGCN[WWW 2019] | ✔️ | ||||
| HetGNN[KDD 2019] | ✔️ | ✔️ | |||
| HeGAN[KDD 2019] | ✔️ | ||||
| HGAT[EMNLP 2019] | Short text classification | ||||
| GTN[NeurIPS 2019] & fastGTN | ✔️ | ||||
| RSHN[ICDM 2019] | ✔️ | ✔️ | |||
| GATNE-T[KDD 2019] | ✔️ | ||||
| DMGI[AAAI 2020] | ✔️ | ||||
| MAGNN[WWW 2020] | ✔️ | ||||
| HGT[WWW 2020] | Heterogeneous transformer | ||||
| CompGCN[ICLR 2020] | ✔️ | ✔️ | |||
| NSHE[IJCAI 2020] | ✔️ | ||||
| NARS[arxiv] | ✔️ | ||||
| MHNF[arxiv] | ✔️ | ||||
| HGSL[AAAI 2021] | ✔️ | ||||
| HGNN-AC[WWW 2021] | ✔️ | ||||
| HeCo[KDD 2021] | ✔️ | ||||
| SimpleHGN[KDD 2021] | ✔️ | ||||
| HPN[TKDE 2021] | ✔️ | ✔️ | |||
| RHGNN[arxiv] | ✔️ | ||||
| HDE[ICDM 2021] | ✔️ | ||||
| HetSANN[AAAI 2020] | ✔️ | ||||
| ieHGCN[TKDE 2021] | ✔️ | ||||
| KTN[NeurIPS 2022] | ✔️ | ||||
| HGDL[NeurIPS 2024] | ✔️ | v0.9 | |||
| HGEN[IJCAI 2025] | ✔️ | v0.9 | |||
| HGSketch[SIGIR 2025] | v0.9, graph-level representation / graph classification pipeline | ||||
| HGOT[ICML 2025] | ✔️ | v0.9 | |||
| RMR[KDD 2024] | ✔️ | v0.9 | |||
| HERO[ICLR 2024] | ✔️ | v0.9, heterogeneous version | |||
| HERO_homo[ICLR 2024] | ✔️ | v0.9, homogeneous version | |||
| SEHTGNN[NeurIPS 2025] | ✔️ | ✔️ | ✔️ | v0.9 | |
| HTGformer[SIGIR 2025] | ✔️ | ✔️ | ✔️ | v0.9 | |
| HCMGNN[IJCAI 2024] | ✔️ | v0.9 | |||
| RelGT[arXiv 2025] | ✔️ | v0.9, RelBench task |
| Item | Count | Note |
|---|---|---|
| New model contributions | 10 | Counted by paper/model contribution |
| New registered model names | 11 | HERO and HERO_homo are registered separately |
| Current registered model names | 83 | Based on openhgnn.models.SUPPORTED_MODELS |
| Current registered task/flow entries | 17 | Based on openhgnn.tasks.SUPPORTED_TASKS |
v0.9 model contributions: HGDL, HGEN, HGSketch, HGOT, RMR, HERO/HERO_homo, SEHTGNN, HTGformer, HCMGNN, and RelGT.
For v0.9 model reproduction entries, tasks, datasets, and remaining documentation notes, see the v0.9 release notes and the model reproduction guide.
- Heterogeneous Graph Attention Networks for Semi-supervised Short Text Classification[EMNLP 2019]
- Heterogeneous Information Network Embedding with Adversarial Disentangler[TKDE 2021]
OpenHGNN Team[GAMMA LAB], DGL Team and Peng Cheng Laboratory.
See more in CONTRIBUTING.
If you use OpenHGNN in a scientific publication, we would appreciate citations to the following paper:
@inproceedings{han2022openhgnn,
title={OpenHGNN: An Open Source Toolkit for Heterogeneous Graph Neural Network},
author={Hui Han, Tianyu Zhao, Cheng Yang, Hongyi Zhang, Yaoqi Liu, Xiao Wang, Chuan Shi},
booktitle={CIKM},
year={2022}
}