When Aaron Schildkrout co-founded HowAboutWe.com, he envisioned it to be something very different than the other online dating sites, which relied on complex mathematical calculations to decide whether people were compatible.
On HowAboutWe, users suggest real-life dates that they would like to go on — seeing a movie at a local theater or taking a walk, for example — and interested users respond through the site’s messaging system. “It’s about getting offline, going to the real world and getting chemistry,” Schildkrout said. “We branded ourselves as the offline dating site, as explicitly an alternative to these profile-heavy matching algorithm dating sites.”
But since its founding in 2009, HowAboutWe has evolved to depend more on formulas, not less. “With hundreds of thousands of people we could show you, who we show you has become increasingly important,” Schildkrout said. “To create a great experience, we need to get smarter about who we show you.”
It’s how most major dating sites work — processing large volumes of information about which two users are likely to show interest in each other, and tweaking the algorithms as more data on successes become available. But depending on the target demographic and the site’s philosophy on what makes a good match, the methodologies vary widely.
HowAboutWe’s two-person data science team, for instance, created an algorithm that combines a user’s profile information, such as date ideas and demographics, with data about a person’s behavior on the site, such as what profiles he or she looked at and how often.
Compared with casual daters on HowAboutWe, eHarmony believes its users are looking for something different: long-term relationships. As a result, people are required to fill out a personality questionnaire with hundreds of parts. Based on decades of data about thousands of happily married couples, eHarmony then predicts which users are likely to be compatible.
“It’s not the sexy big-data stuff of 2012 or 2013, it’s good old-fashioned social research using samples. At this point, we’ve probably looked at 50,000 married couples over the years,” said eHarmony data scientist Steve Carter. The research allows the company to make educated guesses: “What if those two people are married — how happy would they be?”
Historical data is paired with the constraints people place on their matches — target age range, for instance — and the probability, based on the users’ previous behavior and matches, that the two will find each other interesting enough to contact each other. The site’s data team includes a handful of psychologists and computer scientists as well as about 10 developers.
Match.com President Amarnath Thombre says what users often say they are looking for is not always the kinds of profiles they actually view.
For instance, some of the people who said they didn’t want to date people with kids were contacting people who had them. “We said, ‘We’re going to base these things way more on actions you take. . . . If you start breaking your rules, we’re going to start ignoring your rules,’ ” he said.
Through most of its 18-year history, Match.com’s algorithm has drawn on historical data and personality-based compatibility predictions, but it implemented a user-action-driven algorithm in 2009.
AshleyMadison.com, a matchmaking site for people searching for affairs outside their relationships, doesn’t have much historical data to draw upon and is mostly algorithm driven, founder Noel Biderman said.
“At least when it comes to the topic of infidelity, traditional research avenues have been kind of absent,” he said. “There aren’t a lot of universities out there that can give you wholesale data on how unfaithful this population or society is or what triggers this.”
The site’s algorithm is frequently adjusted to reflect aggregate user behavior and individual preferences, Biderman explained.
When the site launched in 2002, for instance, married women invariably searched for married men, but today, about 12 percent are interested in single partners. And search results today are more likely to reflect more-diverse preferences. The data team also observed that the rate of interracial interaction on AshleyMadison far outpaced that of traditional in-person dating and adjusted the algorithm to match people of different ethnicities.
Although automated algorithms provide possible matches, true love still relies on something less tangible.
Former Washington resident Laura Frederick joined Match.com two years ago after a bad break-up, having heard about the site from a friend. In her first month, she received several messages, but none struck her as genuine — most were too creepy or too cheesy, she said.
Finally, she got a message she liked, from Dan Wade. She had shown up among Wade’s top matches for the day, and he sent her a note mentioning details from her profile, such as her love for Chicago-style hot dogs and her support for the Chicago Bears.
Frederick was intrigued. After they exchanged three messages online, “I felt it would be a bit more genuine to meet in person,” she said. They got a drink on the Georgetown waterfront.
Two years later — two weeks ago — Frederick and Wade were married.
“As in the real world, first impressions can sometimes be everything,” she said. “Dan pretty much nailed it.”