eHarmony Blog.com

Hearts Day usually means dozens and dozens of romance stories on the news. This year, online dating is in every newspaper. I’m going through several news items today to find the gems, like this one:

Robert Mitchell got some technical details wrong, but hey, what can anyone expect from ComputerWorld?

Online dating under the covers: How Joe got the girl – Computerworld Blogs, dated 13 February 2009.

• STEP 1: Joe signed on and filled out the profile. To participate, he had to fill out a 400-question demographic and psychological profile questionnaire. Thoughtful responses are the key to success, he says. “Honesty really is what makes the filtering work,” he says. To that end, he not only tried to be extremely honest, but had two friends review his answers. eHarmony then scored the profile and used proprietary algorithms to identify matches which were sent out to him in small batches.
• STEP 2: After receiving his first group of matches, Joe reviewed them, found someone of interest and sent a request to initiate communication. The subject either accepted or declined the invitation.
• STEP 3: If both parties agreed to communicate, Joe selected three canned questions (from a list of about 40) to send to the subject. Each question included four possible answers.
• STEP 4: The recipient responded with any of the generic answers, could answer her own way – or could end the communication at that point.
• STEP 5: After receiving a response, Joe had to decide whether to end communication or continue with another question.
• STEP 6: Once all questions were answered, the process reversed. The woman either selected three questions or could create up to three of her own to send them, one at a time, to Joe, who would then answer.
• STEP 7: If both parties wanted to continue they then opened a line of communication using an eHarmony-hosted messaging function that allows posting back and forth.
• STEP 8: If all went well, Joe and the prospective date then could choose to speak using the e-Harmony teleconference number. Joe didn’t use this. “Most people exchange e-mail addresses during STEP 7, and later phone numbers, he says.
• STEP 9: Finally! Joe set up an in-person date.
  STEP 10: Joe repeated the process, circling back at various steps over and over again.

Mitchell’s blog entry is actually a supplement to a longer CW article Online dating: The technology behind the attraction, dated 13 February 2009. The article has a lot of key information in it — that a lot of cool things have upgraded under the hood in the dating site in the last six months.

How It Is Made — eHarmony Matches

(Cue Discovery Channel music here.)

The technology that powers these dating sites ranges from incredibly simple to incredibly complicated. Unsurprisingly, eHarmony has one of the most sophisticated data centers. Joseph Essas, vice president of technology [bio], says the company stores 4 terabytes of data on some 20 million registered users, each of whom has filled out a 400-question psychological profile (eHarmony’s founder is a clinical psychologist).

The company uses proprietary algorithms to score that data against 29 “dimensions of compatibility” — such as values, personality styles, attitudes and interests — and match up customers with the best possible prospects for a long-term relationship.

A giant Oracle 10G database spits out a few preliminary candidates immediately after a user signs up, to prime the pump, but the real matching work happens later, after eHarmony’s system scores and matches up answers to hundreds of questions from thousands of users. The process requires just under 1 billion calculations that are processed in a giant batch operation each day. These MapReduce operations [definition] execute in parallel on hundreds of computers and are orchestrated using software written to the open-source Hadoop software platform [definition].

Once matches are sent to users, the users’ actions and outcomes are fed back into the model for the next day’s calculations. For example, if a customer clicked on many matches that were at the outset of his or her geographical range — say, 25 miles away — the system would assume distance wasn’t a deal-breaker and next offer more matches that were just a bit farther away.

“Our biggest challenge is the amount of data that we have to constantly score, move, apply and serve to people, and that is fluid,” Essas says. To that end, the architecture is designed to scale quickly to meet growth and demand peaks around major holidays. The highest demand comes just before Valentine’s Day. “Our demand doubles, if not quadruples,” Essas says.

How eHarmony tugs heartstrings into subscribing

… eHarmony, which has the most comprehensive user profiles, may be the most sophisticated in the ways in which it leverages that information. It pulls information — more than a terabyte of data each day — from its Oracle database into high-performance Netezza data warehouse appliances that slice and dice users into behavioral and demographic “buckets.”

“We use [Netezza] to do a lot of offline calculations to try to understand patterns and business intelligence about user behavior,” explains Essas. Some of that feeds back into the matching process, but it also helps eHarmony persuade users to subscribe to its service. “Because we know more about them, we can target them much better,” says Essas. Messaging is tailored to each user’s behavior on the site — and their personality type. …

We cited a press release on Netezza last January 28.

How eHarmony polices the site

… eHarmony has recruited outside help to combat the problem. In addition to in-house tools, Essas says, the company has contracted with Iovation Inc., which offers ReputationManager, a service that gathers information on individuals’ illicit activity from online dating and other sites and makes it available to [companies who subscribed to ReputationManager's service]. …

Wow, this is the very first time I’ve heard that eHarmony uses Iovation. Read what one of our readers, David Evan, wrote about it in February last year.

In a nutshell, Iovation blacklists computers (fingerprinted by a combination of IP address, cookies, software versions, useragent, etc.) who commit online fraud or do “negative reputation” activities.

“Do the matching algorithms produce better matches that lead to long-term relationships?”

… Dan Ariely [bio] doesn’t think so. “The sites are claiming a lot, but show no evidence of doing anything useful in terms of matches,” says Ariely, a professor of behavioral economics at MIT who is researching ways in which online dating sites can do a better job.

Ariely hasn’t examined how well those proprietary matching algorithms work, since eHarmony and other sites won’t release the details. But he suspects that they’re not very effective. “My unsupported guess is that their algorithms are placebos,” he says. …

Well, Ariely, clearly it’s a sophisticated placebo.

Next in my reading list: Online dating: Your profile’s long, scary shelf life