LightGCN on Pytorch

This is a implementation of LightGCN (Paper in arXiv) neural net from SIGIR 2020

Supported datasets:

• gowalla
• brightkite

Use prepare_<dataset_name>_dataset.py for download and splitting by time

Supported models:

• iALS is matrix factorization model from implicit open-source library
• TopNModel recommends top items from all user feedback
• TopNPersonalized recommends top items from unique user feedback
• TopNNearestModel recommends nearest by last user location items (domain-specific for geo features)
• LightGCN
• Catboost fitting with LogLoss/YetiRank and ranking candidates

Training:

Main script is train.py which trains model from MODEL setting in config.yaml file

Also there is fit_catboost.py script which trains catboost ranking model