Musings about the Netflix Prize

The Netflix Prize is a contest by the movie distribution company Netflix. The goal is to come up with the best algorithm for predicting a user's future preferences in movies based on the ratings he or she has given to previously viewed movies. I have a really good idea for this but am too lazy and busy to invest the time in trying it out. So I'm posting it here.

First of all, I would state what I imagine is a typical and very poor movie rating system. If one rates a handful of movies, the algorithm decides which genre is your favorite. Then it suggests to you the most popular movies from your favorite genres. Lame lame lame LAME!

My idea involves concentrating on what I believe are the most important ratings for determining a user's actual heart in movies. These ratings are when a user has very different likes for different movies in a trilogy or a series that are nominally the same. Suppose someone loved Star Wars Episode Three (5 stars) but hated Star Wars Episode Two (1 star). For the lame algorithm described previously, it would merely average 1 and 5 stars to get 3 stars for the user's general preference for the science fiction genre. I say that the few places where one really sees how a user feels about film are shown when they have unexpectedly different ratings for nominally similar films. Supppose you love Dr. No but hate From Russia With Love, suppose you love the Joel episodes of Mystery Science Theatre but hate the Mike and Mrs. Forrester ones. That says everything about your tastes. My algorithm would ignore most of the ratings except these differentiating ones from series, and go seek out other users who had done the same. Then my algo would serve the user with films which were 5-star rated from these similar users.

That I believe is a winning algorithm.