# RFHND

**Repository Path**: zmy-ovo/rfhnd

## Basic Information

- **Project Name**: RFHND
- **Description**: Tackling Over-smoothing on Hypergraphs: A Ricci Flow-guided Neural Diffusion Approach
- **Primary Language**: Python
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-01-06
- **Last Updated**: 2026-01-13

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Tackling Over-smoothing on Hypergraphs: A Ricci Flow-guided Neural Diffusion Approach
This repository contains the official implementation of the **RFHND** framework, a hypergraph neural network inspired by  Ordinary differential equations (ODEs). RFHND generalizes message passing to hypergraphs using a diffusion-based approach.


## Features
- ODE-inspired message passing for hypergraphs  
- Evaluated on **11 hypergraph benchmark datasets**  
- Simple training pipeline with dropout, batch normalization, and weight decay settings


##  Datasets
We evaluate RFHND on the following benchmark datasets:
### Academic Hypergraph Datasets
- `CocitationCora`
- `CocitationCiteseer`
- `CocitationPubmed`
- `CoauthorshipCora`
- `CoauthorshipDBLP`
### Real-World Hypergraph Datasets
- `Zoo`
- `NTU2012`
- `ModelNet40`
- `Walmart` 
- `Senate`
- `House`


## Install dependencies:
pip install -r requirements.txt


## Run the Code
### For the following Datasets  
(`CocitationCora`, `CocitationCiteseer`, `CoauthorshipCora`, `CoauthorshipDBLP`)
python -u main.py --dataset <dataname> --lr 0.001 --weight_decay 0.001 --input_dropout 0.1 --device 0
Replace <dataname> with one of the supported datasets listed above.

### For other Datasets
python -u main.py --lr 0.001 --weight_decay 0.001 --input_dropout 0.1 --device 0
The data for these datasets will be loaded from the ./data/ directory.

#learning-rate, weight decay, dropout rate are individually tuned for each dataset based on validation performance.

## Contact
For questions or collaborations, please contact: zhoumengyao@amss.ac.cn