04-02-08 - 4

I find the Netflix Recommender kind of useless. When I'm browsing around movies trying to find new things, I don't really want rating *correlation* I want movie *similarity*. Like if I'm browsing arty Japanese movies, I don't want it to be telling me "people who liked this also liked" the documentary on Herzog and Kinski, or some damn Ingmar Bergman movie, which seems to always correlate to the stuff I like regardless of the original movie. I want recommendation of other arty Japanese movies. And in fact there's a lot of merit to the graphical representation of something like GNOD, though gnod is a broken ass piece of crap. I'd like to see the nearby similar movies in a 3d space, with the axes meaning something, (the 3 most prominent SVD vectors would make good axes).

Basically a good film-buff human is way better at recommending movies that are related to something I like. That really shouldn't be. The computer has way way more information about my taste history and about the movies that are out there, but it still isn't close to making good recommendations.

No comments:

old rants