I have been using Netflix for several years now, and while I believe it to be one of the most usable and cleanly designed websites on the interent, its algorithm to recommend movies to me has fallen way short of my expectations. I have rated roughly 600 movies on Netflix, which is more than enough to give them a good idea of what I'm in to...yet nearly all the movies that they recommend to me I dislike.
The ratings they display to you are based on how you rate other movies...which means that "The Divinci Code" could be rated as 5 stars to me, yet 3 stars to you based on how you (and others with similar ratings as you) rate other movies. This seems like a really smart way to build a rating system, but it sure doesn't work for me. Personally I would rather view overall ratings of movies rather than an algorithmic one...that way I know I'm looking at real ratings vs. ratings a computer generated for me.
Netflix must be aware of this flaw, because, according to the New York Times, they've started a contest offering a million dollar prize to anyone who can improve their rating system by at least 10% (hopefully they'll use a better system to measure improvement than they use for ratings).
The winning solution (if there is one) will not only be useful to Netflix, but useful for all recommendation algorithms out there...and just may result in pushing the internet to the next level.