LCFN

Codes for papers:

1. Wenhui Yu and Zheng Qin. 2020. Graph Convolutional Network for Recommendation with Low-pass Collaborative Filters. In ICML.

2. Wenhui Yu, Zixin Zhang, and Zheng Qin. 2022. Low-pass Graph Convolutional Network for Recommendation. In AAAI

3. Wenhui Yu, Xiao Lin, Jinfei Liu, Junfeng Ge, Wenwu Ou, and Zheng Qin. 2021. Self-propagation Graph Neural Network for Recommendation. In TKDE.

This project is for our model LCFN and baselines.

• Environment: Python 3.6.8 :: Anaconda, Inc.
• Libraries: tensorflow 1.12.0 numpy 1.16.4 scipy 0.18.1 pandas 0.18.1 openpyxl 2.3.2 xlrd 1.0.0 xlutils 2.0.0

Please follow the steps below:

1. Pretraining 1.1 Run _hypergraph_embeddings.py in folder pretraining. 1.2 Run _graph_embeddings.py in folder pretraining. 1.3 Run _main.py in folder pretraining (set EMB_DIM as 128, 64, 42, and 32 in p_params.py). We also provide downloading for pretraining: https://drive.google.com/file/d/1UV3KO_5wOKkr4v5FePD19cVGhGifz4SR/view?usp=sharing https://pan.baidu.com/s/1KuI4gHViONl3tzRT1Prv1w (password: 1234) You can choose one of these two URLs for downloading (we recommend the first one). Downloaded and unzip LCFN_dataset.zip, and use it to replace the folder dataset in our project.

2. Run _main.py in our project (datasets, hyperparameters can be set in params.py). 2.1 Tuning models (if you want to change datasets or hyperparameters): We provide a automatic tool to tune models with respect to learning rate \eta and regularization coefficient \lambda. Set "tuning_method" in line 24 in _main.py as 'tuning' and run _main.py. The best \eta and \lambda and corresponding performance can be returned. 2.2 Testing models: Set "tuning_method" as 'test' and run _main.py.

3. Check results in folder experiment_result. Collect results by result_collection.

• In the result_collection folder, we provide a tool for results collection. Please read the manual for details.

• We also release our tuning results in folder supplementary_material.

• In the dataset folder, we prodive Amazon and Movielens to conduct our experiments. For each dataset, we split it to three subsets: train_data, validation_data, and test_data. You can use our processed datasets, or construct them from the raw data by running amazon.py and movielens.py. For the raw data please find on:

1. Amazon http://snap.stanford.edu/data/amazon/productGraph/categoryFiles/reviews_Electronics_5.json.gz

2. Movielens http://grouplens.org/datasetss/movielens/1m

Please cite one of our papers if you use our codes:

@inproceedings{LCFN, title={Graph Convolutional Network for Recommendation with Low-pass Collaborative Filters}, author={Yu, Wenhui and Qin, Zheng}, booktitle = {ICML}, year={2020} }

@inproceedings{LGCN, title={Low-pass Graph Convolutional Network for Recommendation}, author={Yu, Wenhui and Zhang, Zixin and Qin, Zheng}, booktitle={AAAI}, year={2022} }

@article{SGNN, title={Self-propagation Graph Neural Network for Recommendation}, author={Yu, Wenhui and Lin, Xiao and Liu, Jinfei and Ge, Junfeng and Ou, Wenwu and Qin, Zheng}, journal={TKDE}, year={2021} }