LightGCN Recommender Pytorch-Geometric(PyG)/Jupyter notebook implementation

Contact: ztor2k@gmail.com, shlife9056@gmail.com

• This repository includes LightGCN(He et al., 2020) recommender system pytorch-geometric/Jupyter notebook implementation with Python.
• Gowalla dataset from LightGCN paper is used for validation.
• We corrected some minor errors in the original code and improved them more intuitively for the Jupyter notebook environment.
• This code uses test set of given dataset as validation set for simple implementation. In real use, validation set should be extracted from train set separately from test set.

References

docs

• https://arxiv.org/pdf/1905.08108.pdf (Neural Graph Collaborative Filtering; NGCF)
• https://arxiv.org/pdf/2002.02126.pdf (LightGCN)
• https://arxiv.org/ftp/arxiv/papers/1205/1205.2618.pdf (Bayesian Personalized Ranking; BPR)

codes

• https://github.com/sh0416/bpr (BPR implementation)
• https://github.com/gusye1234/LightGCN-PyTorch (Official LightGCN pytorch implementation)
• https://colab.research.google.com/drive/1KKugoFyUdydYC0XRyddcROzfQdMwDcnO?usp=sharing (LightGCN pytorch implementation)