Friday, October 8, 2010

The PLAGUE of recommender systems for the Online Dating Industry

I strongly agree with the paper
"A Survey of Accuracy Evaluation Metrics of Recommendation Tasks"
when it says
" ... consider a recommender system for an online dating site. Precision describes what proportion of the suggested pairings for a user result in matches. The false positive rate describes what proportion of unsuitable candidates are paired with the active user. Since presenting unsuitable candidates can be especially undesirable in this setting, the false positive rate could be the most important factor." page 2945

The first recommender system (for the Online Dating Industry) I saw in a paper was:
"Recommender System for Online Dating Service (2007)"

then I saw the one offered by IntroAnalytics.

I had asked IntroAnalytics about the range of its algorithm.

NOW is coming "The PLAGUE of recommender systems for the Online Dating Industry"

"Reciprocal Recommender System for Online Dating" final version
"Reciprocal Recommender System for Online Dating" demo

"Learning User Preferences in Online Dating"

"AI Dating: Development of a Novel Dating Application with Fuzzy Inferencing"

Those NEW recommender systems are only RUBBISH, useless at all for serious dating proposals.
Many recommender systems do not take into account the discovery uncovered by Eastwick and Finkel 2008; also Kurzban and Weeden, 2007; Todd, Penke, Fasolo, and Lenton, 2007 who found that people often report partner preferences that are not compatible with their choices in real life.

Some online dating sites had been using Behavioural Bidirectional Recommendation Engines for years, like PlentyOfFish, and they could not outperform compatibility Matching Methods based on personality profiling.

There is a range convergence phenomenon between the 3 mains tools online dating sites can offer: searching by your own, Bidirectional Recommendation Engines and Compatibility Matching Methods. Any member receives on average 3 or 4 prospective mates as selected / recommended / compatible for dating purposes per 1,000 (one thousand) members screened in the database.
They all 3 are performing the same for serious daters, with a high percentage of false positives, like gun machines shooting flowers.

That range convergence phenomenon is what I had called "the online dating sound barrier", in 2003, when I had discovered than problem, 7 long years ago.

Latest Research in Theories of Romantic Relationships Development, which outlines: compatibility is all about a high level on personality* similarity* between prospective mates for long term mating with commitment.
*personality measured with a normative test.
*similarity: there are different ways to calculate similarity, it depends on how mathematically is defined.
Without offering the 16PF5 (or similar test measuring exactly the 16 personality factors) for serious dating, it will be impossible to innovate and revolutionize the Online Dating Industry.


  1. Hi, I am a developer for an online dating website, and I have just stumble upon your blog. First I must say that I found it interesting, but I completely disagree with most of your comments, mostly because they are very bias towards your personality match formula.

    In regards to this post, I wonder if you read those papers before calling it a plague. I have read the ones on reciprocal recommenders and your comments towards it sounds like you have only glimpse through the titles but have not read them.

    Your statement "recommender systems do not take into account ... people often report partner preferences that are not compatible with their choices in real life." is the topic of "Learning User Preferences in Online Dating" where they compare what people say to what people do. The first paper uses what people do.

    Obviously no dating website can capture the user's preferences beyond what the user does in the website. Unless, of course, you rely on ad-hoc psychological tests that put all users in the same bag.

    In my opinion, personality matching scores might be great as a selling point, but have no real substance when it comes to improving people matching.

  2. Have you seen the next generation of recommender systems are going to include personality traits?

    Please read the post:

    Personality-based Recommender System