Inside Match.com: It's all about the algorithm.

Stories from the Financial Times. 
July 30 2011 7:12 AM

Inside Match.com

It's all about the algorithm.

(Continued from Page 1)

Yet Thombre says his experience at i2, where he spent years finding ways to move products around the country more efficiently, was perfect preparation for the online dating industry. And once at Match, he, Ginsberg and a team of nine maths ­whizzes hired by Thombre, set about updating the Match algorithm. "The one thing I knew was numbers and analytics, so we started building a numbers team here," he told me. "It's just supply and demand. The same principles work, no matter what kind of numerical problem you're solving."

The way the Match algorithm learns, he says, is similar to the way the human brain learns. "When you give it stimuli, it forms neural pathways," he says. "If you stop liking something, those shut off. It's learning as you go." The same principles are powering the recommendation engines at popular sites around the web. Amazon uses similar ­technology to recommend new products for people to buy, Pandora learns from likes and dislikes to customise its internet radio stations, and Netflix famously offered $1m to anyone who could improve the effectiveness of its algorithm by 10 per cent.

"With Netflix, people are constantly rating movies," Thombre told me. "But there's only one The Godfather, and you rate it once." Predicting preferences in human beings is altogether more complicated. "Even if you like The Godfather, The Godfather doesn't have to like you back," he said. "The whole problem of mutual matching makes the problem 10 times more complicated."

It is a subtle shift, but one with profound implications. "Before, matches were based on the criteria you set. You meet her criteria, and she meets yours, so you're a good match," Thombre explained. "But when we researched the data the whole idea of dissonance came into focus. People were doing something very different from the things they said they wanted on their profile."

Advertisement

As a result, Match began "weighting" variables differently, according to how users behaved. For example, if conservative users were actually looking at profiles of liberals, the algorithm would learn from that and recommend more liberal users to them. Indeed, says Thombre, "the politics one is quite interesting. Conservatives are far more open to reaching out to someone with a different point of view than a liberal is." That is, when it comes to looking for love, conservatives are more open-minded than liberals.

With a mountain of data in its servers from the 75 million users it has had since it was founded, Match has been able to uncover a series of curious trends. Some findings are obvious. Women are less likely to e-mail with men who live far away, men who are older than they are, and men who are short. Other findings are more nuanced. Catholic women are especially unlikely to e-mail a Hindu or atheist male. While men are most particular about hair colour, a woman's income is less important to them. "We are so focused on behaviour rather than stated preferences because we find people break from their stated preferences so often," Thombre says.

The Match algorithm is constantly at work behind the scenes, scouring terabytes of data and working to find possible matches. Likely candidates are suggested when users ask to see "more like this" and are also put forward through the "Daily5", a selection of profiles e-mailed to users each day.

But it is not enough for Match simply to suggest dates without gauging the effectiveness of its efforts. When each Daily5 profile is viewed, the user has to "rate" that profile before he or she can see the next one. The site asks users if they are interested in the suggested match, and gets a reply of "Yes", "No" or "Maybe". Each answer is recorded and logged with the user's profile, becoming another data point for the algorithm to work with.

It's not known how many dates the algorithm has resulted in. Match can't know what happens offline. Yet it is clear that changes to the algorithm orchestrated by Ginsberg and Thombre have had an impact on Match users' engagement with the site. Since the introduction of the improved ­algorithm, the "Yes" ratings on the Daily5 have increased over 100 per cent. More than half the e-mails sent on Match now originate from ­recommended matches (chiefly Daily5). And during the past year users have logged more than 416 million Daily5 ratings. Says Ginsberg: "If we can get more people going out on dates, it will have a profound impact on our success rate."

. . .

On the evening of April 5, 2010, Jonathan Cambry, a muscular ­professional pianist living in Chicago, switched on his ­computer and logged on to Match.com. Cambry, then 28, had joined Match a few months earlier. He had recently ­separated from a girlfriend, but had grown weary of spending time and money trying to find dates at local bars. "That wasn't something I was interested in, and it gets expensive," Cambry explains.

So, for about $20 a month, Cambry maintained an account on Match under the alias "Wrigley-Pianist", where he could browse the online profiles of thousands of women in the Chicago area, and communicate with them through e-mail and instant message. And on this evening, as an ­unseasonably late hailstorm kept most Chicagoans indoors, he was notified by Match that a user named "soubrette1988" was interested in him, having seen his profile in her Daily5.

Viewing the profile, Cambry, who is black, saw a pretty young white woman who lived nearby and seemed to share his interest in music. He sent her a short e-mail to say hello, and within a day received an e-mail from Karrah O'Daniel, an opera singer. Their first date was a flop, but they made it to a second date, and soon Cambry and O'Daniel were getting serious. It turns out they had attended the same music school, but never met there. They took to playing pieces by Franz Liszt together, recording videos that they would post on YouTube. Six months later, he proposed. The two are to wed on October 1 at a church in Minnesota.