This repository is the implementation of our paper's model 'Contrastive Learning of Dual-view User Representation for Sequential Recommendation'.

• data/
  • steam/
    • steam_reviews.json.gz: raw rating file and metadata of Steam dataset;
    • preprocess.py: the data preprocess script;
    • gen_test_neg.py
    • attr_file.py
  • movielens/
    • ratings.dat: raw rating file of Movielens dataset(UserID::MovieID::Rating::Timestamp);
    • movies.dat: genre file of Movielens dataset(MovieID::Title::Genres);
    • preprocess.py: the data preprocess script;
    • gen_test_neg.py
    • attr_file.py
  • toys/
    • Toys_and_Games.json.gz: raw review file of Toys dataset;
    • meta_Toys_and_Games.json.gz: raw metadata of Toys dataset;
    • preprocess.py: the data preprocess script;
    • gen_test_neg.py
    • attr_file.py
• model/
  • main.py
  • config.py
  • ...
• weights/

The code has been tested running under Python 3.6.7, with the following packages installed (along with their dependencies):

• torch == 1.7.1+cu101
• numpy == 1.18.5

Steam, Movielens, Amazon Toys&Games dataset can be respectively downloaded from 'http://cseweb.ucsd.edu/jmcauley/datasets.html#steam_data/' 'https://grouplens.org/datasets/movielens/' 'http://deepyeti.ucsd.edu/jianmo/amazon/index.html'

Please first put data files of Steam, Movielens and Toys into data/steam, data/movielens ans data/toys, then run preprocess.py, gen_test_neg.py, attr_file.py successively to process data preparation. 最终data/steam,data/movielens和data/toys中的数据文件应包含attr_test.txt, attr_train.txt, test_candidate_1_50.npy, test.txt, train.txt。

for convinence, we also upload the preprocessed data

$ cd src
$ python model/main.py

The settings of datasets and parameters can be altered in model/config.py. The model checkpoints and training log will be saved in weights/.