CoNHD: Co-Representation Neural Hypergraph Diffusion

CoNHD is a hypergraph neural network tailored for modeling node features specific to each hyperedge. This repository contains the official implementation of CoNHD. For more details, pleasse check our paper.

🎉 News

• 📢 Our paper "Modeling Edge-Specific Node Features through Co-Representation Neural Hypergraph Diffusion" has been accepted by CIKM 2025 (Full Research Paper Track)!

Getting Started

Clone Repo

git lfs install
GIT_LFS_SKIP_SMUDGE=0 git clone git@github.com:zhengyijia/CoNHD.git
cd CoNHD

Environment Setup

The code has been tested with Python 3.8 and CUDA 11.3. To set up the environment, run the following commands:

conda create -n conhd_env python=3.8
conda activate conhd_env
bash install.sh
pip install -r requirements.txt

Preprocessing

The datasets can be preprocessed using the following commands, which generates the initial node features through hypergraph random walks:

cd script
bash ./preprocess.sh

Training and Evaluation

Run the following commands to train CoNHD on each ENC dataset and obtain the corresponding evaluation results:

cd script/edge_dependent_node_classification
bash ./run_[emailEnron, emailEu, StackOverflowBiology, StackOverflowPhysics, DBLP, AMinerAuthor, cora_outsider, dblp_outsider, citeseer_outsider, pubmed_outsider].sh

Acknowledgements

This project benefits from the following repositories:

• WHATsNet: The general training pipeline is based on the implementation of WHATsNet.

• EDHNN: The implementation of diffusion operators builds upon the EDHNN implementation.

• AllSet: The outsider identification datasets are curated from the corresponding datasets provided by AllSet.

We sincerely appreciate the developers of the above repositories.

Citation

If you find this work useful or use this work in your research, please consider citing our paper:

@inproceedings{zheng2025modeling,
  title={Modeling Edge-Specific Node Features through Co-Representation Neural Hypergraph Diffusion},
  author={Yijia Zheng and Marcel Worring},
  booktitle={Proceedings of the 34th ACM International Conference on Information and Knowledge Management},
  pages={4444--4454},
  year={2025},
  url={https://doi.org/10.1145/3746252.3761094}
}