The seedy side of matchmaking algorithms

It’s scary how much it will affect people. I try to ignore some of it, or I’ll go crazy. We are getting to the point where we have a social responsibility to the world because we have this power to influence it.

Jonathan Badeen, Co-Founder and Chief Strategy Officer, Tinder

Swipe right 100 times! Answer 200 questions to find your perfect date. Download high resolution cat photos. Flex. Humblebrag: The internet is full of tips and tricks to energize your online dating life. But even if you put everything into practice, your chances of landing a date depend on an X factor – matchmaking algorithms.

Online dating sites began experimenting with compatibility matching in the early 2000s. This allowed dating sites to monetize their offerings, boost user engagement, and more. In addition, most of these sites claimed to offer “scientific correspondence” based on user preferences.

eHarmony was the first dating platform to create and patent a matching algorithm in 2000. The idea was to reduce divorce rates by intervening in the mating decisions of the site’s unique users. However, by today’s standards, the original eHarmony algorithm was naive – using a regression-based approach to match users based on variables believed to predict long-term relationship satisfaction.

In 2004, OkCupid added algorithmic matching to its core search functionality. The algorithm assessed compatibility using “matching percentages” extracted from user questions and answers. Then, each question was weighted according to its level of importance (ie “not relevant”, “a little”, “somewhat”, “very”). OkCupid gave users control over the matching process. Users could answer some of the questions and let the algorithm figure out the others.

Today, online dating sites or apps use sophisticated machine learning algorithms to predict user preferences based on implicit feedback.

To sweep up

In 2009, Grindr rolled out collaborative filtering to better understand user preferences. Collaborative filtering makes recommendations based on similar user patterns. The same technique is used to recommend products on Amazon and movies on Netflix.

Then came Tinder and Hinge in 2012.

Tinder’s interface resembles a deck of playing cards, with users swiping left to reject a profile and swiping right to match. Tinder used the Elo system to rate users’ desirability and match them with others in the same league. The Elo system is used in the FIDE (chess) rating to assign a score to players based on their previous wins/losses and the skill level of their opponents.

Ratings work the same way on Tinder, with a swipe to the right of a desirable person having the biggest impact on a user’s score, much like beating a chess grandmaster matters more than beating an amateur. Tinder claims to have removed the Elo rating, but is smug about its new system.

Many believe that Tinder uses something similar to the Gale-Shapley algorithm like Bumble or Hinge: the algorithm collects your online data: Facebook, Instagram, Spotify, screen time, dating app usage, etc to find out who you are, what you like and who is likely to like you.

When browsing the profile cards of your preferred gender, Bumble considers general attractiveness and popularity as the primary factors in determining profile order. So even if you don’t get any matches, you see the most attractive profiles first, which leads to a pleasant experience. The same fun you get from window shopping (think playful loops). However, Bumble has never admitted to using the Elo rating to rank the attractiveness of profiles.

One of the issues with using collaborative filtering for matchmaking is the possibility of gender and racial bias creeping into the algorithms. Instead of making dating more inclusive, collaborative filtering will likely reproduce the same biases seen offline.

Desperate a lot?

Have you ever noticed that you get the most likes in the first few days? Frontloading is by design. It serves two purposes; The first is validation, so you won’t uninstall the app. Second, the algorithm learns your preferences and general attractiveness to activate a feedback loop. The idea is to customize and optimize the algorithm. Interestingly, social casino games use the same technique to keep the customer hooked.

In Bumble, only women can start a conversation with their matches, and the match will expire if the conversation is not initiated within the first 24 hours. Those who have a subscription can see the people who have swiped directly in their “beeline”.

Similarly, Tinder shows that you no longer have profiles in your area. But give it some time, and profiles will start popping up in droves again. Again, this could be a rationing strategy to create demand and keep users coming back for more.

Also, some dating apps use shadowban to prevent users from frequently deleting and creating new profiles. Matchmaking algorithms tap into your primal brain and hold you captive by creating addictive behavior. Hinge’s tagline is: The dating app designed to be deleted. But the algorithm tells a different story.

Lance B. Holton