MAGNN_ini

MAGNN for benchmark

Frist, run trans_format.py to transform our benchmark data to MAGNN format.

Then, run ipython preprocess_LastFM.ipynb (or LastFM_magnn) to get preprocessed data.

Next, run python run_LastFM.py --save-postfix LastFM or (LastFM_magnn) to train model.

Last, run python test_LastFM.py --save-postfix LastFM or (LastFM_magnn) to test model.

MAGNN

This repository provides a reference implementation of MAGNN as described in the paper:

MAGNN: Metapath Aggregated Graph Neural Network for Heterogeneous Graph Embedding.
Xinyu Fu, Jiani Zhang, Ziqiao Meng, Irwin King.
The Web Conference, 2020.

Available at arXiv:2002.01680.

Dependencies

Recent versions of the following packages for Python 3 are required:

• PyTorch 1.2.0
• DGL 0.3.1
• NetworkX 2.3
• scikit-learn 0.21.3
• NumPy 1.17.2
• SciPy 1.3.1

Dependencies for the preprocessing code are not listed here.

Datasets

The preprocessed datasets are available at:

• IMDb - Dropbox
• DBLP - Dropbox
• Last.fm - Dropbox

The GloVe word vectors are obtained from GloVe. Here is the direct link for the version we used in DBLP preprocessing.

Usage

1. Create checkpoint/ and data/preprocessed directories
2. Extract the zip file downloaded from the section above to data/preprocessed
   • E.g., extract the content of IMDB_processed.zip to data/preprocessed/IMDB_processed
3. Execute one of the following three commands from the project home directory:
   • python run_IMDB.py
   • python run_DBLP.py
   • python run_LastFM.py

For more information about the available options of the model, you may check by executing python run_IMDB.py --help

Citing

If you find MAGNN useful in your research, please cite the following paper:

@inproceedings{fu2020magnn,
 title={MAGNN: Metapath Aggregated Graph Neural Network for Heterogeneous Graph Embedding},
 author={Xinyu Fu and Jiani Zhang and Ziqiao Meng and Irwin King},
 booktitle = {WWW},
 year={2020}
}