1. Input format(s): Train/Validate/Test sets have the same format

• For each basket sequence, baskets {b_i} are separated by '|', e.g., b_1|b_2|b_3|...|b_n

• For each basket b_i, items {v_j} are separated by a space ' ', e.g., v_1 v_2 v_3 ... v_m

2. How to train:

• Step 1: Generate pre-computed correlation matrix C using cmatrix_generator.py.
  • The 'nbhop' parameter is to generate the Nth-order correlation matrix
  • The default output directory is "data_dir/adj_matrix"
• Step 2: Train the Beacon model using main_gpu.py ".
  • Support 3 modes: train_mode, prediction_mode, tune_mode
  • The format of the prediction file is as follows: Target:gt_basket|item_candidate_1:score_1|item_candidate_2:score_2|

3. If you find the code useful in your research, please cite:

@inproceedings{le2019beacon,
  title={Correlation-Sensitive Next-Basket Recommendation},
  author={Le, Duc-Trong, Lauw, Hady W and Fang, Yuan},
  booktitle={Proceedings of the International Joint Conference on Artificial Intelligence},
  year={2019},
}

• Python == 3.6
• Tensorflow == 1.14
• scipy.sparse == 1.3.0

@Please drop me an email (ductrong.le.2014 at smu.edu.sg) if you need any clarification. Thanks 👍