The Attraction Algorithm - Mash-Up Week


By Brian Fencil

Photo courtesy of Azlan DuPree.

Since inception, online dating sites have been experimenting with how to match people based on shared interests. Many sites have used long, detailed surveys and then studied the data from thousands of profiles and relationships to perfect their algorithms and test questions. These sites scored answered questions, turned things like values and interests into numbers, as well as compatibility into math equations.

Other sites tried to narrow the field by only appealing to niches in the dating market, such as online dating for farmers or girls with short hair.

All of these sites had a fault that couldn’t be remedied: they could not suggest users based on physical attraction. No dating site or app had figured out how to define a “type” in order to quantify beauty so that a computer could calculate attraction. But now, the dating service Three Day Rule (TDR) uses facial recognition software to teach a computer about a user’s type.

Clients of TDR fill out an in-depth personality assessment, which is used to match them on interests and life goals. To suggest matches that a client might be physically attracted to, TDR uses facial recognition software to analyze photos of their exes.

Though TDR launched in 2010, the company’s recent partnership with has been misreported in mainstream media. News articles carried titles like “Missing Your Ex? Dating Service uses Facial Recognition to Find a Lookalike of Your Previous Partner…” and other iterations of the same idea.

Erin Janzer, who handles press inquiries for TDR, clarifies the partnership with Match for BTR. Janzer explains that TDR is not trying to “find you a clone of your ex” but they use photos of exes because “usually a client’s ex is a good indication of who they are attracted to.”

The original press release from Match and TDR stated the two companies formed a “strategic partnership” that would allow Match users access to TDR’s facial recognition software. In return, support from Match would enable TDR to expand in terms of the company’s team and markets. TDR is currently only available in New York City, Los Angeles, San Francisco, and Chicago but plans to expand into Dallas, Boston, and Washington DC later this year.

The Washington Post tested TDR’s facial recognition software, and found that it is surprisingly accurate. The publication sent TDR a few photos of some men and had them compared to the players in the world cup. The software indentified hair color, face and eye shape, as well as the eyebrow structure of the person in the photos and did the same analysis for World Cup players. The matches it returned are incredibly similar.

TDR is the first company to use facial recognition software for online dating, but it may become more prevalent as the technology develops. As it gains popularity, Tinder and other apps could phase out, since this technology can right or left swipe automatically and sort through thousands of users in an instant. Though, inevitably, there will be users who still prefer to manually sort through matches.

Additionally, this year marks a huge development in facial recognition that could make it easier to apply the software to uses other than security and photo tag suggestions. Facebook unveiled DeepFace, an artificial intelligence system that recognizes faces almost as well as people do. DeepFace creates 3-D models of facial features and accurately identifies faces 97.25 percent of the time (people are accurate 97.53 percent of the time).

The use of facial recognition software might also help reduce our biases in judging profile pictures. Currently, there are many studies that show incidental details–the side of the face shown in a picture and if the person is looking at the camera or away–all affect the amount of attention their profile receives.

If a computer suggests someone based on their appearance and ignores irrelevant details, we might overlook these incidentals and have a less biased eye. Also, users may start posting more passport-style pictures that are easier for facial recognition software to analyze.