Contextualized Fairness for Recommender Systems in Premium Scenarios

Abstract

Previous studies have demonstrated the potential bias and fairness issues in real recommender systems. Fairness issue is generally defined as equality of services between groups. However, fairness does not imply equality for all users/items in real scenarios, as premium users/items, who have paid for their services, are supposed to have better experiences. In this work, we focus on the fairness issue in recommender systems under premium scenario. Firstly, we design a new metric to measure the fairness issue when there are premium users. Then, our analyses demonstrate that premium users/items get better recommendations than standard users/items on average. While our further study shows that by controlling confounding factors, premium users are provided with similar system performances as standard users, which indicates current recommender systems can be unfair. Finally, we propose a flexible and contextual fairness-aware recommendation framework by considering a desired distribution to fit user (or item) groups' scores, and experimental results show that it is a solution to provide better services for premium users/items.