FairJob: A Real-World Dataset for Fairness in Online Systems

Official repository of the paper FairJob: A Real-World Dataset for Fairness in Online Systems (Mariia Vladimirova, Federico Pavone, Eustache Diemert) available on ArXiv (https://arxiv.org/abs/2407.03059).

Data

The dataset FairJob and its detailed description is available at https://huggingface.co/datasets/criteo/FairJob.

Download the data in the subfolder \data before running the code in the repository.

Code structure and examples

The file functions.py implements all the functions and classes used in the experiments. In order to run an example of model fit to FairJob data you can do:

python example_fit.py --dummy=1 --name=EXAMPLE

You can check all available options via:

python example_fit.py --help

If you want to run the experiment of logistic regression based on different randomization of train-test split of the data, you can do:

python example_simulations_LR.py --lr_fair=1 --fair_frac=1.0 --name=EXAMPLE

You can also in this case check all options available for example_simulations_LR.py with the flag --help.

Paper results

In order to reproduce the results reported in the paper, please refer to the executions listed in paper_results.sh and to the notebook dataset_analysis.ipynb for the post-processing of the results.

A Note on License

This code is open-source. We share most of it under the Apache 2.0 License.

Citation

If you use the dataset in your research please cite it using the following Bibtex excerpt:

@article{vladimirova2024fairjob,
  title={{FairJob: A Real-World Dataset for Fairness in Online Systems}},
  author={Vladimirova, Mariia and Pavone, Federico and Diemert, Eustache},
  journal={arXiv preprint arXiv:2407.03059},
  year={2024}
}