Too Cool For School: How Popularity Bias Effects Match Making on Online Dating Platforms
Let’s set the scene: You’re walking down the hall of your school and you see the popular person that “everybody” has a crush on. How could they not? They’re attractive, smart, and witty. You have a crush on them yourself, but you wouldn’t dare try to approach them because of the potential embarrassment of being shut down. Instead, you crush on them from a distance in the hopes they notice you in the same way you notice them. This age-old storyline is something we see in books, movies, TV shows, and even music videos time and time again. In fact, we see this type of interaction, or lack thereof, happen on online dating platforms all the time. A paper was written on research about this phenomenon detailing how popularity bias on online dating platforms affects the company’s problems of matchmaking and revenue maximizing (Celdir et al., 2022).
As online dating platforms shifted from a free-browsing model of online dating to a recommendation model, there has been a rise in matchmaking technology. Popularity value is assigned to every user based on things like attractiveness, employment (high-paying jobs over low-paying ones), education (prestigiousness of institution), etc. Those who are deemed more popular by the algorithm are more likely to be recommended to users as potential partners and will receive more messages hoping to connect. However, popular users are less likely to accept messages creating a matchmaking, revenue-maximizing conundrum. The degree of popularity bias, the idea that people will be recommended popular items at the expense of non-popular items stopping people interested in the nonpopular item from seeing them, benefits the revenue and matchmaking-maximizing objective at the expense of unpopular users. Online dating platforms usually generate revenue through advertisements shown to users as they look at recommended partners. Here we see the importance of popular users, as they are more likely to be recommended during the potential partner search and more time will be spent viewing their profiles, thus increasing the company revenue. However, popular users are less likely to accept messages. Therefore, if they are the only ones being recommended, yet they are not accepting many messages, this seems like it would lessen the likelihood of successful matches. Right?
A two-sided matching game that consists of three stages (recommendation, sending, and accepting/matching) was created to explore the mechanisms behind popularity bias on online dating platforms and how it solves the platform’s problem of revenue and matchmaking-maximizing. The results: a certain degree of bias against unpopular users is necessary for the max-max outcome. The revenue-maximizing objective leads to more time spent exploring popular users’ profiles and more messages being sent to those users in the second stage of the matching game. This helps the matchmaking-maximizing objective of more successful matches if there is a high number of messages being sent in the third stage of the matching game. The bias makes it so the two objectives are not at odds. In comparison to the unbiased objective, where recommendations are done at random, the platform and user have greater success with the popular user-biased algorithm. However, this only works if the popular users are not too selective when it comes to accepting messages, so much so they reach an “unattainable” status.
The value system assigned to users and the value-optimizing agenda of these platforms align with the discussion on matching games. However, there are added layers and nuances that go into finding perfect matches that make constricted sets hard to get rid of. For example, the paper discusses gendered differences in popularity bias and how women are much more selective in comparison to men when it comes to sending and accepting messages. Furthermore, the more popular a woman (or man) may be on the platform, the increased likelihood that they have potential partners outside of the dating platform which increases their selectivity. As a result, the match will not be “perfect” but rather, yield the best outcome. An exploration of popularity bias on online dating platforms is useful in developing prediction models of users’ behaviors and future decisions when it comes to interacting with future partners. Furthermore, the presence of popularity bias in matching games on two-sided matching platform recommendations affects users’ likelihood of finding potential matches not just on online dating platforms, but on ride-sharing platforms, home-rental platforms, gig-worker finder platforms, and more.
Reference:
Celdir, M., Cho, S.-H., & Hwang, E. H. (2022). Popularity bias in online dating platforms: Theory and empirical evidence. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.4053204