Fairness Metrics for Recommender Systems

Abstract

Fairness is a hot topic in recommender system research in recent years. Researchers have resorted to regularization and other techniques to reduce fairness problems. However, a lot of research literature adopts classic evaluation metrics for recommender system results. There has been little attention paid to the fairness metrics for recommender system evaluation. In this paper and for the first time in the research history of recommender systems, we propose a set of fairness metrics based on extreme value theory. In the experiment section, we evaluate different classic algorithms and fair AI technologies with our newly invented fairness metrics.