An examination of the role of relevance in fair exposure-based ranking. For more details please see our SIGIR 2023 paper.
Run the following commands to clone this repo and create the Conda environment:
git clone git@github.com:AparnaB/role-of-relevance-in-fair-ranking.git
cd role-of-relevance-in-fair-ranking/
conda env create -f environment.yml
conda activate rel_exp
We provide Rnormal.csv
and Rpareto.csv
of the synthetic datasets in the data
folder. We also provide notebooks for processing all data in the notebooks/
folder.
To reproduce the experiments in the paper which involve training ranking models using pretrained click and propensity models, run the main.py
file with different dataset and experimental settings.
Sample bash scripts showing the command can be found in bash_scripts/
.
We aggregate results and generate tables using scripts in the lib
folder.
We provide script used simulate fair re-ranking interventions in lib/fair_reranking.py
.
To reproduce the imbalanced groups experiment described in Section 6.3 of the paper, run the lib/get_imbalanced_fairness.py
script.
If you use this code in your research, please cite the following publication:
@article{balagopalan2023,
title={The Role of Relevance in Fair Ranking},
author={Balagopalan, Aparna and Jacobs, Z. Abigail and Biega, Asia},
conference={SIGIR},
year={2023}
}