Leveraging Opposite Gender Interaction Ratio as a Path towards Fairness in Online Dating Recommendations Based on User Sexual Orientation

TL;DR

A novel metric, Opposite Gender Interaction Ratio (OGIR), is proposed, as a way to investigate potential unfairness for users with varying preferences towards the opposite gender, and a fair recommender system based on re-weighting and re-ranking strategies is proposed to mitigate these associated imbalance challenges.

Abstract

Online dating platforms have gained widespread popularity as a means for individuals to seek potential romantic relationships. While recommender systems have been designed to improve the user experience in dating platforms by providing personalized recommendations, increasing concerns about fairness have encouraged the development of fairness-aware recommender systems from various perspectives (e.g., gender and race). However, sexual orientation, which plays a significant role in finding a satisfying relationship, is under-investigated. To fill this crucial gap, we propose a novel metric, Opposite Gender Interaction Ratio (OGIR), as a way to investigate potential unfairness for users with varying preferences towards the opposite gender. We empirically analyze a real online dating dataset and observe existing recommender algorithms could suffer from group unfairness according to OGIR. We further investigate the potential causes for such gaps in recommendation quality, which lead to the challenges of group quantity imbalance and group calibration imbalance. Ultimately, we propose a fair recommender system based on re-weighting and re-ranking strategies to respectively mitigate these associated imbalance challenges. Experimental results demonstrate both strategies improve fairness while their combination achieves the best performance towards maintaining model utility while improving fairness.

Figures (11)

• Figure 1: Dataset analysis (a) gender identity distribution and their average ratings; (b) interaction type distribution and their average ratings; (c) OGIR distribution of female/male users; (d) user counts and average degrees according to OGIR.
• Figure 2: Utility performance of three models on five metrics, where groups are divided based on even width bins for discretizing OGIR into three groups ($G_1=\{u|\text{OGIR}_u\in [0, \frac{1}{3})\}$ with $G2$, $G3$ similarly defined). $G_3$ consistently has better performance.
• Figure 3: Two potential causes of unfairness (a) group quantity imbalance; (b) group calibration imbalance.
• Figure 4: Analysis on the utility and fairness performance impacts associated with the re-weighting hyperparameter $p$.
• Figure 5: The utility and fairness performance of variants (1) the baseline model (Model); (2) the re-weighting model (Model$_\text{rw}$); (3) the re-ranking model (Model$_\text{rr}$); and (4) the re-ranking model based on re-weighted model (Model$_\text{rw\&rr}$).