Finding Your Preferred Seller on Tinder: A Matching Markets Analysis
Nobody said online dating was easy—but by better understanding the matching algorithms that operate behind the screen, it may finally be possible to beat the odds and find your soulmate.
In her recent article, Kamya Pandey takes a deeper look at the matching algorithms behind Tinder, Bumble, and Hinge, perhaps three of the most popular online dating platforms at the moment. While exposing some of the inner-workings of Tinder’s Elo score, Bumble’s match queue, and Hinge’s Gale-Shapley algorithm, Pandey ultimately reaches the conclusion that a particular matching algorithm can only be as effective as the personal information inputted by users is accurate. This post will explore some of the similarities between the algorithms online dating platforms use and the matching market frameworks that can be found in a textbook.
One of the key similarities between standard matching markets and online dating matching algorithms is that both prioritize the creation of pairings where social welfare is maximized. In a textbook example, this could take the form of a buyer of a house maximizing their payoff and the seller maximizing the price (while maintaining buyer interest). For Tinder’s Elo score, this means connecting users that have matching levels of overall desirability, with desirability referring to one’s rate of right swipes (an action taken to indicate interest in a match). Though Tinder also factors in the rate at which users actually message their matches or check the app, the general premise of matching is the same: maximize the total valuation of all matches based on interest/compatibility metrics. One caveat arises when considering the following question: does a socially optimal allocation rely on the users/nodes being able to establish preferences/prices for all possible matches? In the case of a typical matching market, nodes are able to establish their preferences for all available nodes where a match could be formed, allowing for the computation of perfect matchings to occur in full. However, dating apps like Tinder and Hinge impose limits on user swipes/likes to make matches feel more important and scarce. Whether this limitation diminishes the meaning of socially optimal allocation in this context is certainly up to interpretation, but it could be overcome if each app is viewed as having pools of users that form bipartite networks, where overlap between networks is allowed.
The algorithms used on dating apps also strike a similarity to matching markets when it comes to stability of the network. A standard matching market could be viewed as stable when a perfect matching exists between all nodes, no matter the criteria used for matching. Hinge’s use of the Gale-Shapley algorithm for matching takes a similar approach—stability occurs when all users prefer their current match over other potential matches in the dating pool. In reality, creating a perfect matching where all matches actually connect is highly unlikely, but the perspective on match stability occurring when all nodes have one distinct match is still quite similar.
Though the crossover between online dating matching algorithms and matching markets is not perfect, it is interesting to acknowledge some of the obstacles (like match limits and the variability of user preferences) that separate matching frameworks from a textbook and those in the real world.
References:
https://medium.com/swlh/dating-data-an-overview-of-the-algorithm-afb9f0c08e2c#:~:text=Hinge%20uses%20the%20Gale%2DShapley,to%20people%20with%20similar%20preferences
Does Online Dating Work? A Look at the Matching Algorithms of the Most Popular Dating Apps