Top-K Contextual Bandits with Equity of Exposure

Source code for our paper "Top-K Contextual Bandits with Equity of Exposure" published at RecSys 2021.

Acknowledgments

This work relies heavily on the source code and data accompanying the "Carousel Personalization in Music Streaming Apps with Contextual Bandits" paper published at RecSys 2020. We are grateful to the original authors, W. Bendada, G. Salha and T. Bontempelli, for open-sourcing their simulation environment and enabling this work. Their original repository can be found here.

Reproducibility

• Download the data under src/data/, following instructions in the repository for the original RecSys 2020 paper.
• Run the following commands to reproduce the experimental results reported in the manuscript: cd src/; python3 main.py --policies ts-lin-pessimistic-reg-2000,ts-lin-pessimistic-reg-2000-personalised-shuffle-99,ts-lin-pessimistic-reg-2000-personalised-shuffle-975,ts-lin-pessimistic-reg-2000-personalised-shuffle-95,ts-lin-pessimistic-reg-2000-personalised-shuffle-90,ts-lin-pessimistic-reg-2000-shuffle-6 --n_rounds 50 --print_every 1 --gamma 0.9 --output_path results.json.

Paper

If you use our code in your research, please remember to cite our paper:

    @inproceedings{JeunenRecSys2021_B,
      author = {Jeunen, Olivier and Goethals, Bart},
      title = {Top-K Contextual Bandits with Equity of Exposure},
      booktitle = {Proceedings of the 15th ACM Conference on Recommender Systems},
      series = {RecSys '21},
      year = {2021},
      publisher = {ACM},
    }