BasConv

A graph neural network based framework to do the basket recommendation. Our paper is

Liu, Zhiwei, et al. "BasConv: Aggregating Heterogeneous Interactions for Basket Recommendation with Graph Convolutional Neural Network." Proceedings of the 2020 SIAM International Conference on Data Mining. Society for Industrial and Applied Mathematics, 2020.

Data

Part of the instacart dataset is given in this repo. We have three files:

• trainu2b.txt: \ Each row is user_id, basket_id, basket_id, ...
• trainb2i.txt: \ Each row is baset_id, item_id, item_id, ...
• testb2i.txt: \ Each row is baset_id, item_id, item_id, ...

The original data we use in the paper is too big to upload. You may find complete dataset of instacart here

Running

Run the code under ./basConv/ folder with script:

python basConv.py --dataset inscart_1 --regs [1e-4] --alg_type basconv --embed_size 64 --layer_size [64,64] --lr 0.0002 --save_flag 1 --pretrain -1 --batch_size 4096 --epoch 2000 --verbose 50 --node_dropout_flag 0 --mess_dropout [0.2,0.2]

More info regarding the arguements, please refer to the ./basConv/parser.py file.

Enviroment

Python = 3.6\ Tensorflow = 1.8+ \ Numpy, Scipy, scikit-learn should be installed accordingly.

Reference

@inproceedings{liu2020basconv,
  title={BasConv: Aggregating Heterogeneous Interactions for Basket Recommendation with Graph Convolutional Neural Network},
  author={Liu, Zhiwei and Wan, Mengting and Guo, Stephen and Achan, Kannan and Yu, Philip S},
  booktitle={Proceedings of the 2020 SIAM International Conference on Data Mining},
  pages={64--72},
  year={2020},
  organization={SIAM}
}

Acknowledgement

We reuse some part of the code in Neural Graph Collaborative Filtering https://github.com/xiangwang1223/neural_graph_collaborative_filtering