Collaborative filtering on dating sites

Out of purely technological interest (obviously) I’ve been researching some dating sites recently. One feature I discovered most of them have is a kind of ‘short-list’ of people you like. You look around and add the profiles you find appealing onto a list so you can access them easily when you’re aiming for a next victim. Sometimes, the subjects in question are aware of their presence on your short-list, sometimes not. In any way they consist of links between people, links they’ve chosen to add themselves.

So I was thinking, do any of these sites use all this aggregated “X-likes-Y” information to suggest users new profiles to take a look at, basically like Amazon’s ‘people who like Zero7 also like Air’. None of the sites I know has it. Via Google I found a service with ‘reciprocal collaborative filtering’, but honestly, they remind me too much of all the ‘Rate-my-[bodypart].com’ sites.

Some further googling brought me to a discussion at pdesigner from 1996(!) about why dating did not fit into classic collaborative filtering schemes:

If we take the hetero-sexual example of guys adding girls to their hotlists, I can see two ways of using collaborative filtering:

  1. INDIVIDUAL: if you add girl X, and other guys have added X too, what other girls have these guys added, that you might find appealing too? This kind of calculation could be done instantly, but there might be undesired side effects on an individual basis. If a girl convinces 10 guys to add her to their list, and also add all girls that look like Pamela Anderson, she might be presented to anyone who adds one of the Pamelas, even if she is an intelligent, medium-chested natural blond, and as such completely different. Once people get a feel of how individual decisions influence the system, they will try to manipulate it.
  2. CLUSTERED: you could group girls into clusters: a more-or-less homogeneous group of persons that are similar (in age, hair colour, interests, chest size, or a combination hereof). It may sound demeaning, but these profiling cluster techniques have been used on people in direct marketing for years.
    If a guy seems to tap mainly into the ‘blond cheerleaders that like Ricky Martin and Brad Pitt’ group, you could present him with girls from the same category. There might be people that are ‘unclusterable’, but the results would be harder to manipulate through individual choices, since they are based on much more data from more people. The clustering is also not done in real-time, it’s typically updated weekly/monthly (lots of calculations).

Since collaborative filtering started around 1994, and people were discussing using it in dating networks in 1996 (cf. article in Infoworld), why hasn’t it been used on dating sites? It’s just the on-line version of your buddy telling you ‘Hey, I’ll introduce you to this girl, you’ll like her, she’s got piercings!’ , right?

I can think of a couple of reasons people might object:

Basically, apart from techophobia and privacy paranoia, I see no valid reason not to use collaborative filtering for dating. So,, my favourites are Meg Ryan, Christy Turlington and Penelope Cruz! Fetch!

[Listening to: “Haven't you heard” – Patrice Rushen – Sampled Vol 4 (CD 2/2)]
💬 Web2.0