Fairness on Online Platforms
Abstract
Many online platforms, ranging from online retail stores to social media platforms to job search sites, employ algorithms to optimize their offered assortment of items (e.g., products, contents, and jobs). These algorithms tend to prioritize the platforms# short-term goals by solely featuring items with the highest popularity or revenue. However, this practice can then lead to too little visibility for the rest of the items, making them leave the platform, and in turn hurting the platform#s long-term goals. Motivated by that, we will study a fair assortment planning problem, which requires any two items with similar quality/merits to be offered similar visibility. Studying this problem can be of interest to several online platforms, such as online retail stores, job search sites, social media platforms, and movie/music recommendation sites, to name a few. In all of these online platforms, a "winner-take-all" and the "rich-get-richer" phenomena can become prevalent due to the wide adoption of algorithmic recommendations and focus on short-term gains.In this research, we present new (parametric) notions of pairwise fairness that can help the platforms to manage the priceof enforcing fairness. Given our fairness notion, we will develop fair recommendation algorithms that balance efficiency and fairness. Designing such recommendation algorithms is challenging because (i) the problem of designing recommendation algorithms is combinatorial with an exponentially large decision set and (ii) the decisions are governed by operational constraints (e.g., diversity andinventory constraints). To design fair recommendation algorithms, we consider a parameterized notion of fairness that is based on parity of pairwise visibility/exposure across items. Under our fairness notion, we study the platform#s recommendation problems in both offline and online settings, where in the offline setting, the platform has an estimate of the quality score of the items and in the online setting, the platform is not aware of the quality scores and aims to learn them over time while being fair to the items. We willpresent algorithms with provable performance guarantees under both settings and evaluate them in real-world-inspired settings.In the offline setting, our algorithms will (i) rely on approximation separation oracles (that we will design) for the platform#s dual problem, and (ii) achieve robustness concerning estimator errors in quality constraints by considering "chance fairness constraints" that ensure fairness constraints are satisfied with high probability. The designed algorithms for the offline setting will randomize over a small number of high-quality assortments to satisfy the fairness constraints. In the online setting, we imitate our designed algorithms for the offline setting over time while balancing exploration and exploitation. This enables us to obtain balanced data across all the items and alleviate the rich-get-richer phenomenon. Overall, this research provides a very flexible framework topromote fairness across a wide range of platforms, and sheds light on the price of fairness and the fundamental trade-off between fairness and efficiency.This abstract is marked for public release
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 15, 2023
- Source ID
- N000142312584
Entities
People
- Negin Golrezaei
Organizations
- Massachusetts Institute of Technology
- Office of Naval Research
- United States Navy