This repository holds the source code for our paper "Pessimistic Reward Models for Off-Policy Learning in Recommendation" published at RecSys 2021. This work has been extended as "Pessimistic Decision-Making for Recommender Systems", published in the ACM Transactions on Recommender Systems (ToRS) journal.
- Install RecoGym requirements as described in the official repository.
- Install utilities and common packages: TQDM, NumPy/SciPy, Pandas.
- Run
python3 src/PessimisticRidgeRegression.py
, which will run all experiments and dump raw measurements in .csv files, along with coarse visualisation in .png files.
If you use our code in your research, please remember to cite our work:
@article{JeunenTORS_2022,
author = {Jeunen, Olivier and Goethals, Bart},
title = {Pessimistic Decision-Making for Recommender Systems},
journal = {Transactions on Recommender Systems (TORS)},
year = {2022},
publisher = {ACM},
}
@inproceedings{JeunenRecSys2021_A,
author = {Jeunen, Olivier and Goethals, Bart},
title = {Pessimistic Reward Models for Off-Policy Learning in Recommendation},
booktitle = {Proceedings of the 15th ACM Conference on Recommender Systems},
series = {RecSys '21},
year = {2021},
publisher = {ACM},
}