This is the Tensorflow implementation by @LIU-YUXI for SelfGNN model proposed in this paper:

Self-Supervised Graph Neural Networks for Sequential Recommendation
Yuxi Liu, Lianghao Xia, Chao Huang* SIGIR2024

* denotes corresponding author

In this paper, we propose a novel framework called Self-Supervised Graph Neural Network (SelfGNN) for sequential recommendation. The SelfGNN framework encodes short-term graphs based on time intervals and utilizes Graph Neural Networks (GNNs) to learn short-term collaborative relationships. It captures long-term user and item representations at multiple granularity levels through interval fusion and instance modeling. Importantly, our personalized self-augmented learning structure enhances model robustness by mitigating noise in short-term graphs based on long-term user interests and personal stability.

You can run the following command to download the codes faster:

git clone https://github.com/HKUDS/SA-GNN.git

Then run the following commands to create a conda environment:

conda create -y -n sagnn python=3.6.12
conda activate sagnn
pip install matplotlib==3.5.1
pip install numpy==1.21.5
pip install scipy==1.7.3
pip install tensorflow_gpu==1.14.0

We utilized four public datasets to evaluate: Gowalla, MovieLens,Yelp and Amazon. Following the common settings of implicit feedback, if user  has rated item , then the element  is set as 1, otherwise 0. We filtered out users and items with too few interactions.

We employ the most recent interaction as the test set, the penultimate interaction as the validation set, and the remaining interactions in the user behavior sequence as the training data.

The datasets are in the ./Dataset folder:

- ./Dataset/amazon(yelp/movielens/gowalla)
|--- sequence    # user behavior sequences (List)
|--- test_dict    # test item for each users (Dict)
|--- trn_mat_time    # user-item graphs in different periods (sparse matrix)
|--- tst_int    # users to be test (List)

The original data of our dataset can be found from following links (thanks to their work):

• Yelp: https://www.yelp.com/dataset
• Amazon-book: https://cseweb.ucsd.edu/~jmcauley/datasets/amazon/links.html
• Gowalla: SNAP: Network datasets: Gowalla (stanford.edu)
• Movielens: MovieLens 10M Dataset | GroupLens

If you want to process your data into the several data files required for SA-GNN (i.e., sequence,test_dict,trn_mat_time,tst_int), you can refer to the following code for preprocessing the raw data of Amazon-book:

1. Download the original data file (for example, amazon_ratings_Books.csv from links for original datasets) and run preprocess_to_trnmat.ipynb to get the trn_mat_time and tst_int files, as well as other intermediate files (train.csv,test.csv).
2. Run preprocess_to_sequece.ipynb, which reads in the intermediate files (train.csv and test.csv) and finally generates the sequence and test_dict files.

You are welcome to modify the preprocessing code as needed to suit your data.

You need to create the ./History/ and the ./Models/ directories. The command to train SA-GNN on the Gowalla/MovieLens/Amazon/Yelp dataset is as follows.

• Gowalla

./gowalla.sh > ./gowalla.log 

• MovieLens

./movielens.sh > ./movielens.log 

• Amazon

./amazon.sh > ./amazon.log 

• Yelp

./yelp.sh > ./yelp.log

If you find this work is helpful to your research, please consider citing our paper:

@article{liu2024selfgnn,
  title={SelfGNN: Self-Supervised Graph Neural Networks for Sequential Recommendation},
  author={Liu, Yuxi and Xia, Lianghao and Huang, Chao},
  journal={arXiv preprint arXiv:2405.20878},
  year={2024}
}

Thanks for your interest in our work!