This repo is a tf implementation of "Contrastive State Augmentations for Reinforcement Learning-Based Recommender Systems" (SIGIR 2023). If you have any question, please open an issue or contact [email protected].

• cuda 11.2
• Python 3.7
• Tensorflow-gpu 2.6.0
• pandas
• numpy

We provide a runnable version of RC15 and RetailRocket. You can run this program directly using dataset in "Google Drive". You should download and move it under "./data/."

If you want to get "meituan" dataset and more processing details, please contact [email protected].

Make sure the following packages are installed with the correct version.

conda install tensorflow-gpu==2.6.0
conda install pandas
conda install numpy
pip install trfl
pip install tensorflow-probability==0.14.0 

The following commands can be used to train and evaluate CSA based on GRU4Rec:

cd code/GRU4REC/
python CSA_N.py

You need to modify the directory of dataset and set hyperparameters for different base models and data according to the data provided in the paper.

@article{ren2023contrastive,
  title={Contrastive State Augmentations for Reinforcement Learning-Based Recommender Systems},
  author={Ren, Zhaochun and Huang, Na and Wang, Yidan and Ren, Pengjie and Ma, Jun and Lei, Jiahuan and Shi, Xinlei and Luo, Hengliang and Jose, Joemon M and Xin, Xin},
  journal={arXiv preprint arXiv:2305.11081},
  year={2023}
}