This repository includes the implementation for User-oriented Fairness in Recommendation: Yunqi Li, Hanxiong Chen, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang. 2021. User-oriented Fairness in Recommendation. In Proceedings of the Web Conference 2021 (WWW'21).

For inquiries contact Yunqi Li ([email protected]) or Hanxiong Chen ([email protected]) or Yongfeng Zhang ([email protected])

@inproceedings{li2021user,
  title={User-oriented Fairness in Recommendation},
  author={Yunqi Li, Hanxiong Chen, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang},
  booktitle={Proceedings of the the Web Conference 2021},
  year={2021}
}

Python 3.6.6

Packages:

pandas==0.24.2
Gurobi==9.0.2

Gurobi is a commercial optimization solver. To run our code, please first install Gurobi and purchase a license. Without a license or fail to install properly, our code will not be able to run.

• The processed datasets are in ./dataset/
• Amazon Datasets: The origin dataset can be found here.
• For each dataset directory contains processed splitted testing datasets for re-ranking.
  • 0.05_count_*_test_ratings.txt: grouping by total number of interactions.
  • sum_0.05_price_*_test_ratings.txt: grouping by total consumption.
  • max_0.05_price_*_test_ratings.txt: grouping by maximum price.

• To run the code, please put the ranking file generated by recommendation model under the corresponding dataset folder. For example, to run model with 5Beauty-rand dataset, put the "*_rank.csv" file under "dataset/5Beauty-rand/" directory.
• Ranking csv file format: uid \t iid \t score \t label
  • uid: user id column
  • iid: item id column
  • score: predicted score column
  • label: 0 or 1 to indicate this is a negative sample or positive sample

• Before running the code, please update the info of "src/model.py" in "__main__" section. You need to update dataset_name, model_name, group_name_title, group_1_file, group_2_file.
  • group_name_title: select from "0.05" (interaction), "max_0.05"(max price) and "sum_0.05"(total consumption)
  • group_1_file: select from "_count_active_test_ratings.txt" (interaction), "_price_active_test_ratings.txt" (max price/total consumption)
  • group_2_file: select from "_count_inactive_test_ratings.txt" (interaction), "_price_inactive_test_ratings.txt" (max price/total consumption)
• You can chage the variable epsilon to control the strictness for fairness.

> cd user-fairness/src/
> python model.py

The result file will be located in ./results/ by default.