Improved Rank Aggregation Under Fairness Constraint
Authors: Diptarka Chakraborty, Himika Das, Sanjana Dey, Alvin Hong Yao Yan
IJCAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We complement our theoretical guarantee by performing extensive experiments on various real-world datasets to establish the effectiveness of our algorithm further by comparing it with the performance of state-of-the-art algorithms. |
| Researcher Affiliation | Academia | Diptarka Chakraborty1 , Himika Das2 , Sanjana Dey3 and Alvin Hong Yao Yan1 1National University of Singapore 2TU Wien 3UMONS EMAIL, EMAIL, EMAIL, EMAIL |
| Pseudocode | Yes | Algorithm 1 COLORFUL BI-PARTITION |
| Open Source Code | Yes | The data and code is available on Github3. 3https://github.com/Aussiroth/Improved-Fair-Rank-Aggregation |
| Open Datasets | Yes | We use two datasets introduced in previous work studying fairness in rank aggregation. The first dataset from [Kuhlman and Rundensteiner, 2020] is taken from a fantasy sports website for American football... The second dataset from [Wei et al., 2022] contains the rankings of 7 users over 268 movies. |
| Dataset Splits | No | The paper uses datasets but does not explicitly provide training/test/validation dataset splits, their percentages, or specific counts needed for reproduction. |
| Hardware Specification | Yes | The implementation is in Python 3.12 and experiments were performed on a laptop running Windows 11 using a Ryzen 6800HS processor and 16GB of RAM. |
| Software Dependencies | Yes | The implementation is in Python 3.12 and experiments were performed on a laptop running Windows 11 using a Ryzen 6800HS processor and 16GB of RAM. We also use Integer Linear Programming (ILP) to find the optimal solution where possible for comparison. The Integer Linear Programs are implemented with CVXPY [Diamond and Boyd, 2016], using SCIP [Bolusani et al., 2024] as the solver. |
| Experiment Setup | Yes | For all experiments, the values of αi, βi are selected to be equal to the proportion of elements belonging to group i in the dataset. This is a natural option, as it maintains the proportion of elements from each group in the top-k. |