I will be analyzing the recommender system behind tinder, one of today’s most popular dating apps. Their target users are people who are single and currently looking for partners. The users goals are to find partners and or dates with people they are interested in and attracted to. Tinder can aid in users accomplishing these goals by recommending people who have aligned interests, who are geographically relevant (e.g. not 300 miles away), and are likely to have some sort of bond. Tinder solves the problem of mutual attraction by only matching people who both agree that they are attracted to the other. Tinder also has some issues with bot accounts, and needs to reckon with that by attempting not to recommend bot accounts.
It is worth mentioning that the goals of Tinder as an organization may not be aligned with the goals of the user. Tinder monetizes by charging a monthly fee to extend the features of their platform. This means that Tinder theoretically loses out when a user finds a partner and no longer needs to pay Tinder to use their features. This leads to two interesting conclusions: Tinder is incentivized to make the base features promising yet ineffective so that users pay for the increased features. Tinder is incentivized to make the overall experience promising yet ineffective so that users do not actually find partners and therefore stop paying Tinder for the service. In this way, Tinder’s goals are not really aligned with the user’s goals and this may be reflected in their recommendation engine.
According Sean Rad, the tinder founder, part of the tinder algorithm is a machine learning system that gives users an Elo score which ranks them by “desirability”. You are only shown profiles of others with similar desirability, meaning if you have a lower desirability you will not even be shown profiles of higher desirability. There are other algorithms such as collaborative filtering and proprietary systems, meaning that the Tinder algorithm functions as an ensemble predicter. The algorithms also actively deny matches at times, even if both users swipe right. Tinder explains this as “giving people with lower rankings a chance”, but it is easy for me to be skeptical and think that this is more Tinder protecting their interests and trying to keep users on the app for a longer period of time.
Another issue mentioned in an article by Diggit is that since the tinder algorithms are based on analyzing swipes and the preferences of people on the app, they may be propogating racial bias in dating. If the general populace on average is biased away from a certain race due to racism, the algorithm will learn that this group is “undesirable” and will be less likely to recommend people of this group or only recommend them to other undesirable people. Studies by OKCupid have confirmed that there is racial bias in dating, and this likely needs to be addressed in any fair recommender system.
Going forward, I believe that Tinder should be more open with their users about the algorithm and how the system is generating matches. This is unlikely to happen as this would potentially hurt their monetization. I also believe that Tinder should try to adjust their algorithms to help fight against biases in the general population, as companies should do their best not to propogate racist or classist idealogies. Finally, I believe that the algorithm should include some variability that allows people to at least view profiles from other levels of “desirability” in order to allow users to make informed choices about who they would like to match with. You can’t accept a match if it is never shown to you!
Rolle, Magdalena. “The Biases we feed to Tinder algorithms.” https://www.diggitmagazine.com/articles/biases-we-feed-tinder-algorithms. Diggit Magazine. 2 Feb. 2019.
Heath, Alex. “You have a secret ‘desirability score’ on Tinder that controls who you see in the app.” https://www.businessinsider.com/tinders-desirability-score-controls-who-you-see-2016-1. Business Insider. 12 Jan. 2016.
Brown, Ashley. “Least Desirable? How Racial Discrimination Plays Out In Online Dating”. https://www.npr.org/2018/01/09/575352051/least-desirable-how-racial-discrimination-plays-out-in-online-dating. NPR. 9 Jan. 2018.