HyperIMBA

Code for "Hyperbolic Geometric Graph Representation Learning for Hierarchy-imbalance Node Classification".

Overview

• main.py: the core of our model, including the structure and the process of training.
• calculator.py: the code about calculating Poincare embedding and class-aware Ricci curvature weights
• dataloader.py: providing data loading and processing.
• models/: including the backbones we used in the main model.

Environment

Our experimental environments are listed in environments.yaml, you can create a virtual environment with conda and run the following order.

conda env create -f environments.yaml

Install

Enter the virtual environment and run the requirements.txt.

pip install -r requirements.txt

Datasets

All the datasets are provided by pytorch_geometric.

Usage

Run the following order to train our model.

python main.py

Reference

@inproceedings{fu2023hyperimba,
  title={Hyperbolic Geometric Graph Representation Learning for Hierarchy-imbalance Node Classification},
  author={Fu Xingcheng, Wei Yuecen, Sun Qingyun, Yuan Haonan, Wu Jia, Peng Hao and Li Jianxin},
  booktitle={Proceedings of the 2022 World Wide Web Conference},
  year={2022}
}