jennyzhang0215 / STAR-GCN

STAR-GCN: Stacked and Reconstructed Graph Convolutional Networks for Recommender Systems

STAR-GCN

The MXNet implementation of STAR-GCN: Stacked and Reconstructed Graph Convolutional Networks for Recommender Systems in IJCAI 2019

Prerequisite

Requirements

• MXNet Version 1.5.0
  • Please build from source
  • Compile the customized segment operator kernel according to README
• GraphSampler
  • Compile the graphsampler API according to README
• GluonNLP

pip install gluonnlp

• spaCy

pip install spacy
python -m spacy download en

Installation

Install the mxgraph python package

python setup.py develop

Training

The training scripts are in the experiments directory. The chosen hyperparameters are stored in yml files in the cfg dirctory.

For example, to train the MovieLens-100k in the transductive setting, we can run

cd experiments/cfg
python ../STAR-GCN.py --ctx gpu0 --cfg transductive_ml_100k.yml

Cite

Please cite our paper if you use this code in your own work:

@inproceedings{zhang2019star,
  title     = {STAR-GCN: Stacked and Reconstructed Graph Convolutional Networks for Recommender Systems},
  author    = {Zhang, Jiani and Shi, Xingjian and Zhao, Shenglin and King, Irwin},
  booktitle = {The 28th International Joint Conference on Artificial Intelligence},
  pages     = {4264--4270},
  year      = {2019}
}