A Pure Keras Implementation of KGCN

This is a pure keras implementation of Knowledge Graph Convolution Network (KGCN) as described in the paper Wang et.al. Knowledge Graph Convolution Networks for Recommender Systems. WWW2019. Author's tensorflow implementation is available in this repo.

Environments

• python==3.6.6
• Keras==2.3.0
• tensorflow-gpu==1.13.1

Run

Prepare data

All the data are copied from author's repo. The rating file of MovieLens-20M is still needed to download first:

wget http://files.grouplens.org/datasets/movielens/ml-20m.zip
unzip ml-20m.zip
mv ml-20m/ratings.csv raw_data/movie/

Preprocess

python3 preprocess.py

The processed data will be stored in data dir.

Train

python3 main.py

Performance

Reference

• Knowledge Graph Convolution Networks for Recommender System, WWW2019
• hwwang55/KGCN