Product Recommendations: Free Beats Fee
David’s right. This whole recomendation-engine thing is mind-boggling.
Amazon’s telling me to buy Pink Floyd because I told them I own a lot of VU. Huh? They’re convinced my predilection for Patti Smith makes “The Concert for Bangladesh” a must-own (the mind reels). Given the vintage of my tunes, I obviously didn’t buy a lot of them on Amazon. But I have been putting their engine to the test by spending way too much time telling Amazon what I already have — to little avail.
Over at GreenCine, it’s much clearer why I get the recomendations I do. But they’re still wrong, wrong, wrong.
Why? In a word, segmentation. On-the-fly tagging is much nimbler than baked-in metadata.
I overwhelmingly rent Asian films, primarily Japanse and Korean (OK, I have obscure tastes — that’s why I bailed from Netflix). GreenCine obviously takes this preference for “asian” into consideration as Asian films account for a preponderance of my personal recomendations. But (and this is a very big but), nearly all the recommendations are anime. This completely disregards the fact my rental history consists of zero anime.
I can click “not interested” on anime suggestions until I get calluses, but there’s always more where that came from. The engine just doesn’t get it. GreenCine segments their anime into no less than 11 subcategories (who knew?), all distinct from a completely separate Animation category. This means one broad and 11 subcategories I’ve never rented DVDs from constitute about 90 percent of my recommendations — presumably because of metadata I’m not seeing; “japanese” and “asian” are likely guesses.
It’s so much cheaper and so much easier for these e-commerce players to go the CGM route. I love Amazon’s customer reviews, lists, and “people who looked at this product actually bought…” links. I’ve rented a ton of movies off Greencine’s excellent, eccentric and often, arcane, user recommendation lists (like Roger Ebert is a Big, Fat, Idiot or Movies that make you want to take a shower).
Building an e-commerce site? Considering a recommendation engine? Save your money. Let your customers do that work for you.