xiaohanhan1019 / CAN

Capturing Multi-granularity Interests with Capsule Attentive Network for Sequential Recommendation

This code is used to reproduce the main experiment of our paper.

Requirements

• Python 3.7.10
• Pytorch 1.8.0

Datasets

The original dataset can be downloaded from:

• ml-1m: https://grouplens.org/datasets/movielens/1m/
• Amazon Books: http://jmcauley.ucsd.edu/data/amazon/
• ml-10m: https://grouplens.org/datasets/movielens/10m/

The format of the preprocessed dataset is the same as RecBole which can be downloaded from:

• ml-1m: https://drive.google.com/drive/folders/1OkDVEqetvOrtbuWebxl4y1JlZ_YjjfWj
• Amazon Books: https://drive.google.com/drive/folders/1FUAScpZtCmArqQS0xdrIqpFx5F7u3TaY
• ml-10m: https://drive.google.com/drive/folders/1OkDVEqetvOrtbuWebxl4y1JlZ_YjjfWj

The *.inter file should be stored in the ./raw_data folder

Code

• ./models folder contains all the baselines and our model CAN
• ./data folder is used to store the processed data

Training

• First you need to process the data by running./raw_data/preprocess.py, make sure the raw dataset is downloaded and stored in the ./raw_data folder
• After processing the data, you can train a model by running ./trainer.py
  • The model's configuration can be modified in ./utils/configs.py
• ./ray_tune.py for tuning the hyper-parameters
  • The model's configuration can be modified in ./utils/tune_configs.py