dhg-bench 0.1.0

An open and unified pipeline for Deep Hypergraph Learning (DHGL)

---

Overview of the Benchmark • Installation • Quick Start • Algorithm References • Citation

DHG-Bench

DHG-Bench is an open and unified pipline for Deep Hypergraph Learning (DHGL) based on PyTorch and PyTorch Geometric. We embark on 17 state-of-the-art hypergraph neural network (HNN) algorithms in node-level, edge-level, and graph-level tasks and analyze their performance in 22 distinct hypergraph datasets.

📔 Overview of the Benchmark

Figure 1: Pipeline of the DHG-Bench.

DHG-Bench provides a fair and comprehensive platform to evaluate existing HNN algorithms and facilitate future DHGL research. Our main contributions are summarized as follows:

• DHG-Bench enables a fair and unified comparison among 17 state-of-the-art HNN algorithms and 22 diverse hypergraph datasets covering node-level, edge-level, and graph-level tasks. To the best of our knowledge, this is the first comprehensive benchmark for deep hypergraph learning.

• We conduct a systematic analysis of HNN methods from various dimensions, including effec- tiveness, efficiency, robustness, and fairness. Through extensive experiments, we reveal both the strengths and limitations of existing algorithms, offering valuable insights to inform and inspire future research in this field.

• We release DHG-Bench, an easy-to-use open-source benchmark library to support future DHGL research. Besides, with our toolkit, users can readily evaluate their algorithms or datasets with less effort.

📦 Installation

Follow the steps below to install and configure DHG-Bench properly for your local environment.

Installation for Local Development:

git clone https://github.com/Coco-Hut/DHG-Bench.git 
cd DHG-Bench

Download Datasets

We include all benchmark datasets in the data.zip archive. Users can simply extract the archive with

unzip data.zip

to obtain the data directory. The project structure should look like the following:

DHG-Bench
  ├── data
  │   ├── fair_data
  │   ├── hete_data
  |   ├── hgcls_data
  │   ├── trad_data
  |   └── ...
  └── dhgbench
  └── main.py
  └── parameter_parser.py

Required Dependencies:

DHG-Bench needs the following requirements to be satisfied beforehand:

• Python>=3.9.21
• Pytorch>=2.2.2
• torch_geometric>=2.6.1
• torch-cluster>=1.6.3
• torch-scatter>=2.1.2
• torch-sparse>=0.6.18
• torch-spline-conv>=1.2.2
• deeprobust==0.2.11
• ipdb==0.13.13
• numpy==1.24.3

🚀 Quick Start

The following commands show how to quickly run an existing HNN algorithm across different hypergraph tasks.

For example, to run the HGNN method on the Cora dataset for a node classification task, use the following command:

python main.py --dname=cora --task_type=node_cls --method=HGNN --is_default=True

For example, to run the EDHNN method on the Pubmed dataset for a hyperedge prediction task, use the following command:

python main.py --dname=pubmed --task_type=edge_pred --method=EDHNN --is_default=True

For example, to run the TFHNN method on the stream_player dataset for a hypergraph classification task, use the following command:

python main.py --dname=stream_player --task_type=hg_cls --method=TFHNN --is_default=True

Note that The is_default parameter indicates whether to use the model’s default configuration. If set to False, the model will instead load parameter settings specific to the given dataset. All model parameter files are provided in the lib_yamls directory.

🧩 Algorithm References

ID	Method	Paper	Venue
1	HGNN	Hypergraph neural networks	AAAI 2019
2	HyperGCN	HyperGCN: A New Method of Training Graph Convolutional Networks on Hypergraphs	NeurIPS 2019
3	HNHN	HNHN: Hypergraph Networks with Hyperedge Neurons	ICML WS 2020
4	HCHA	Hypergraph convolution and hypergraph attention	PR 2020
5	UniGNN	UniGNN: a Unified Framework for Graph and Hypergraph Neural Networks	IJCAI 2021
6	AllSetformer	You are AllSet: A Multiset Function Framework for Hypergraph Neural Networks	ICLR 2022
7	HyperND	Nonlinear Feature Diffusion on Hypergraphs	ICML 2022
8	EHNN	Equivariant hypergraph neural networks	ECCV 2022
9	LEGCN	Semi-supervised Hypergraph Node Classification on Hypergraph Line Expansion	CIKM 2022
10	ED-HNN	Equivariant Hypergraph Diffusion Neural Operators	ICLR 2023
11	PhenomNN	From Hypergraph Energy Functions to Hypergraph Neural Networks	ICML 2023
12	SheafHyperGNN	Sheaf Hypergraph Networks	NeurIPS 2023
13	HJRL	Hypergraph Joint Representation Learning for Hypervertices and Hyperedges via Cross Expansion	AAAI 2024
14	DPHGNN	DPHGNN: A Dual Perspective Hypergraph Neural Networks	KDD 2024
15	T-HyperGNNs	T-HyperGNNs: Hypergraph Neural Networks via Tensor Representations	TNNLS 2024
16	HyperGT	Hypergraph Transformer for Semi-Supervised Classification	ICASSP 2024
17	TF-HNN	Training-Free Message Passing for Learning on Hypergraphs	ICLR 2025

📖 Citation

Our paper on this benchmark has be released at: http://arxiv.org/abs/2508.12244

If you use our benchmark in your works, we would appreciate citations to the paper:

@article{li2025dhg,
  title={DHG-Bench: A Comprehensive Benchmark for Deep Hypergraph Learning},
  author={Li, Fan and Wang, Xiaoyang and Zhang, Wenjie and Zhang, Ying and Lin, Xuemin},
  journal={arXiv preprint arXiv:2508.12244},
  year={2025}
}