Fairer-Together-Mitigating-Disparate-Exposure-in-Kemeny-Aggregation

Corresponding source code and experimental implementation for "Fairer Together: Mitigating Disparate Exposure in Kemeny Aggregation" paper in FAcct'23.

EPIK and EPIRA code:

Please see src folder for python scripts for EPIK and EPIRA implementations. Epik is in epik.py and EPIRA is in epira.py,

The baseline method PRE-FE (preprocessing) is in the preprocess_kem.py script.

Comparative baselines:

All baselines used are in the baselines folder.

• KEM (Kemeny). Please see kemeny.py.
• RAPF. Please see baseline_weietal.py.
• PFAIR-KEM. Please see baseline_cacheletal.py.

Experiments:

Mallows datasets.

See the Mallows_Datasets directory.

All experiments utilizing the Mallows datasets are performed in the run_mallows.py script, with results written to the mallows_results.csv file. This includes running all proposed algorithms, comparative baselines, EPIRA with the Copeland rule without the WiG property, and running EPIRA with different voting rules. Each Mallows preference profile is in its own csv file of the format R_disp_<disperasion param>_fairp_<reference ranking>_.csv.

Plotting heatmaps.

heatmaps_mallows.R produces the heatmaps of the consensus accuracy and exposure ratio from each method. It's figures are written to the Mallows_Datasets\plots directory.

Preflib datasets.

See the Preflib_Datasets directory.

All experiments utilizing the preflib datasets are performed in the exp_preflib.py script, with results written to <dataset name>_results.csv files. This includes running all proposed algorithms, comparative baselines, and running EPIRA with different voting rules.

CSRankings.

See the CSRankings_Datasets directory.

All experiments utilizing the cs rankings dataset are performed in the run_csrankings.py script, with results written to <dataset name>_results.csv files. This includes running all proposed algorithms, comparative baselines, and running EPIRA with different voting rules.

Gamma Parameter.

See the Mallows_Datasets directory.

All experiments utilizing the Mallows datasets are performed in the gamma_values.py script, with results written to the mallows_gamma_results.csv file. .

Helper Functions:

Functions to calculate consensus accuracy and exposure ratio are in the metric directory.