There are more than eight million people in New York City, and a myriad of ways to meet them. Given our increasing reliance on social media to stay connected with the people we already know, it’s no surprise that more and more of us are forgoing bars, clubs, sports teams and happy hours, and instead turning to the web to find love.

Some 800,000 New Yorkers now have accounts on OkCupid, a Manhattan-based dating website founded in 2004. These users are counting on four Ivy League math majors (OkCupid’s founders) to find their soul mates—or at least someone to spend a fun Friday night with. I sat down with one of the company’s founders, Christian Rudder, as well as the site’s Chief Technology Officer, Tom Quisel, to talk about how the OkCupid matching algorithm works. Rudder and Quisel revealed some of the invisible controls used to create a better user experience, as well as what their massive cache of data can—and can’t—tell you about your prospective dates.Quisel starts by explaining how New Yorkers’ famously snobbish ideas about distance have led the site’s creators to rethink matchmaking here in NYC.

Tom Quisel of OkCupid
Tom Quisel of OkCupid

Quisel starts by explaining how New Yorkers’ famously snobbish ideas about distance have led the site’s creators to rethink matchmaking here in NYC.

While the user interface of OkCupid feels pretty simple—type up a profile, add some flattering photos, answer some questions and then search away—the infrastructure behind it is much more complex. Rudder details how the site’s matching system works, and Quisel tackles one of the most difficult identity questions currently facing programmers.

Short of asking users to fill out an exit survey after every date, it can be difficult to gauge just how well the matching system is working. The tech team checks the health of their algorithm by looking at what they call “three-ways.” And no, it’s not what you think.

Christian Rudder of OkCupd
Christian Rudder of OkCupd

Many users with profiles on OkCupid have received flattering emails telling them they have been deemed attractive by other OkCupid users. Due to your high attractiveness rating, the email says, you are now going to see more pretty people when you search for matches. So how exactly does the algorithm decide who is good looking?

I have had an OkCupid profile on and off for quite some time. Occasionally I will run into real-life friends on the site. I’ve always thought this is a good sign—I know and like these people already, so the algorithm must be doing something right. Except, of course, that I don’t want to date any of these friends, for one reason or another. This left me wondering: With all of that question-and-answer data, what can’t the algorithm predict about my potential matches and me?

*  *  *

Kat Vecchio is a Brooklyn based documentary filmmaker whose first feature, This Is How I Roll, a film about the quirky beginnings of men’s roller derby, debuted last spring.  She is currently developing a new film on the ramifications of our use of algorithms and data collection to solve subjective problems.  You can find her on Twitter @katvecchio.

Jessica Bal hails from a two-stoplight town in Massachusetts and now resides in a city with too many lights to count, where she produces media for an arts education organization and looks for any excuse to write, photograph and film stories that she’s curious about.

All music by Jahzzar