MetaChat is an informal place for MeFites to touch base and post, discuss and
chatter about topics that may not belong on MetaFilter. Questions? Check the FAQ. Please note: This is important.
05 June 2010
I totally do not understand the Amazon.com algorithms sometimes.→[More:]Today, they said:
"Because you liked 'Christmas with Sonos Handbell Ensemble', you may also like 'Performance' by Mel Torme." More than 6 degrees of separation, perhaps...
Every day for the past 5 days Amazon has sent me notices about Photoshop CS5. OK, I admit that I checked out the upgrade price once (turns out I'm not eligible for the upgrade as I only have CS). But really, I don't need to be notified every day. </end unrelated amazon rant>
This mystifies me constantly! Like I read a historical novel recently and rated in three stars. Suddenly all of these books about America's history started getting recommended to me (like a biography of Betsy Ross [?????]), based on that lukewarm rating--and the book wasn't even set in America!
Their system is almost entirely based on a simple "people who bought X also bought Y," with few refinements. More complex algorithms which take into account all of your ratings (and comparing with other customers) are too computationally time-consuming to be economically feasible, given the huge number of products and users.
Even last.fm, which does much more interesting analysis to generate recommendations and neighbors (similar users), simplified their system a few years ago because it was becoming unmanageable as the userbase grew. But with that change there was a noticeable decline in the quality of matches.
Amazon recommends several books I worked on, but only since I bought an extra copy of one of the other books I did.