My favorite programming assignment in school was a Scheme implementation of the "stable marriage algorithm".
"Let's assume, for the sake of an argument, that you are a village matchmaker given the task of marrying
100 men and 100 women. Each of the men has ranked the women from 1 to 100 in the order of his preference;
each woman, not to be outdone, has similarly ranked the 100 men...Rather than insist that everyone get their first choice, you are instead charged with creating 100 stable
marriages. A set of marriages is said to be stable if there exists no man m and woman w such that m likes
w better than his wife, and w likes m better than her husband. The notion is called "stability" as m's and
w's marriages are unstable, since (albeit sordid and tawdry) m and w could optimize their sorry lot in life
by leaving their mates and running off together."
eHarmony isn't successful because of the matching algorithm. eHarmony is successful because of the steep barriers to entry, perception of safety and privacy, its ability to attract women to the site, the rate at which they regulate match delivery, and the guided communication process.
The result is that you don't waste much time browsing profiles. You don't waste time trying to filter out spammers. You don't waste time coming up with clever pick-up lines. You don't get pestered by matches you close.
The matching algorithm definitely matters, but it's mostly a red herring. If I were eHarmony, I'd be far more worried about my terrible user interface than the matching algorithm.
I perceive that the biggest barrier to entry is the amount of money that you have to spend in marketing. We have been able to match most of their functionality but we don't have nor want to spent that much money in marketing. We haven't been able to come up with alternative approaches.
I just want to make sure its clear that in my previous post, "barrier to join" is probably more accurate than "barrier to entry." I was talking about the amount of time, effort, and money it takes to actually sign up and create a profile. It's a relatively enjoyable process, but it weeds out people who don't have patience.
How many people will just go along with whatever adversity they encounter in the relationship because they've "found their soul mate". It has to be, the computer said so.
The netflix contest was a spectacular failure. I can beat it easily just by using better metrics for the input data.
Really, the contest should have been about finding the data that is most representative of the expected results. But no, they want to throw resources at problems that are unsolvable with the provided data.
e-harmony is very similar to my start up. We operate in Mexico somoscompatibles.com (we are compatible). Our approach is to find personalities and rate how similar they are. I tried eharmony for several months and I don't think they really have a great algorithm, they just make it easier for you to know some people that are somewhat similar to you.
Last year I did a lot of work under contract to build a machine learning based data web site where users would vote on people and the ML part would build up a model of what they were looking for. Unfortunatly, the site has not yet released because of trouble finding a good Javascript person to work on bits of the UI.
By having an exclusionary policy, eHarmony appears to be attempting to improve their results by keeping people out that are less likely to fit their algorithm.
I'm impressed that they've hired so many PhDs, but on the other hand, I don't believe that having the computer match you is going to be very successful overall.
does eHarmony still refuse to match a huge chunk of people? i've heard about 20% or so are unmatched or something like that. which interestingly almost coincides with the commonness of a certain range of personality types in the Jungian theories
speaking of which, i'm surprised Jungian theories aren't more popular. there are specific type pairings which go very well together -- like the people you meet that you just "click" with. well, that clicking can be seen clearly in the theories by the way the types are described
It seems that the more accepted personality classification is the big five. We have thought about denying access to those that score high on neurotisism. Maybe cut off the top ten percent thus protecting our members.
eHarmony's ability to market itself/word-of-mouth is impressive. My relatives/parents are like broken record players 'You know, Roy met his wife through eHarmony...' My friend Mike met his wife through eH as well - I remember him complaining about how long it took to answer the questionnaire set.
"Let's assume, for the sake of an argument, that you are a village matchmaker given the task of marrying 100 men and 100 women. Each of the men has ranked the women from 1 to 100 in the order of his preference; each woman, not to be outdone, has similarly ranked the 100 men...Rather than insist that everyone get their first choice, you are instead charged with creating 100 stable marriages. A set of marriages is said to be stable if there exists no man m and woman w such that m likes w better than his wife, and w likes m better than her husband. The notion is called "stability" as m's and w's marriages are unstable, since (albeit sordid and tawdry) m and w could optimize their sorry lot in life by leaving their mates and running off together."
Non-PDF link: http://64.233.169.104/search?q=cache:abZ7sCKHUmYJ:www.cs.bra...
Not the same as algorithms discussed in the article but fun and interesting.