- Here presents the code of RL-ISN. As the dataset is too big for GitHub, we upload two datasets (i.e., Hvideo and Hamazon) on Bitbucket: https://bitbucket.org/jinyuz1996/rl-isn-data/src/main/. You should download them before training the RL-ISN. The code is attached to our paper: " Reinforcement Learning-enhanced Shared-account Cross-domain Sequential Recommendation" (TKDE 2022) ". If you want to use our codes and datasets in your research, please cite our paper. Note that, our paper is still in the state of 'Early Access', if you want to see the preview version (on IEEE Xplore), please visit: https://ieeexplore.ieee.org/document/9802739.
- python 3.6
- tensorflow 1.12.0
- scipy 1.5.3
- numpy 1.16.0
- pandas 0.22.0
- matplotlib 3.3.4
- Keras 1.0.7
- tqdm 4.60.0
- the definition of Basic recommender see: RL_ISN/ISN_recommender.py
- the definition of Agent see: RL_ISN/ISN_agent.py
- the definition of Enviroment see: RL_ISN/ISN_environment.py
- the definition of Training process see: RL_ISN/ISN_train.py
- the definition of Evaluating process see: RL_ISN/ISN_evaluation.py
- the preprocess of dataset see: RL_ISN/ISN_configuration.py
- the hyper-parameters of RL-ISN see: RL_ISN/ISN_parameters.py
- to run the training method see: RL_ISN/ISN_main.py and the training log printer was defined in: RL_ISN/ISN_printer.py
- The directory named Checkpoint is used to save the trained recommenders ,agents and joint-training models.
- Note that, for different dataset (i.e., HVIDEO and HAMAZON), we set two different settings for the haper-parameters as 'ISN_parameters_Hamazon.py' and 'ISN_parameters_Hvideo.py'