JasonZhangzy1757 / Heterogeneous-Graph-Attention-Network-HAN-PyTorch

Heterogeneous-Graph-Attention-Network-HAN-PyTorch

This is a proof of concept of HAN implementation using pytorch framework. The authors' original code can be found here.

If you find this work helpful for your research, you could cite the original paper as the following:

@inproceedings{wang2019heterogeneous,
  title={Heterogeneous graph attention network},
  author={Wang, Xiao and Ji, Houye and Shi, Chuan and Wang, Bai and Ye, Yanfang and Cui, Peng and Yu, Philip S},
  booktitle={The world wide web conference},
  pages={2022--2032},
  year={2019}
}

This implementation is also inspired by the dgl implemenation and the earlier pytorch implementation of GAT by Diego999. The data processing is copied from ZZy979. Please also check his other brilliant works in his page.

Usage

1. Download the data here and put the data under a new data directory. If you don't have access to Google Drive, you could also check ZZy979's work.
2. python main.py for reproducing HAN's work.
3. python RGCN_baseline.py for adding on an RGCN baseline.
4. Use --dataset to specify the dataset you hope to run against. Currently it's DBLP by default. The options could be: ACM or IMDB.

Performance