RSS
Twitter
Did Someone Finally Win The Netflix Prize?
from the nice-work dept
For years, we've been fascinated by the Netflix Prize -- the $1 million offer to any team that can come up with a recommendation algorithm that is shown to be 10% better that Netflix's current recommendation engine. For years, different teams worked on the problem, and the early improvements were fast, but then progress seemed to stall out. Some different approaches were tested out which pushed the numbers up even further, but getting that last little bit has proven quite elusive... until now. Apparently, two of the leading teams combined efforts and have submitted an entry that breaks the 10% barrier for the first time. If the results are verified, then other teams have 30 days to submit an algorithm that performs even better. But, if they can't, then this team should win the $1 million... at which point they're supposed to tell the world how they did it. Seems like a much more innovation friendly approach than locking it up with a patent.
Reader Comments
(Flattened / Threaded)
Yay!
The unnoticed nugget here is that Netflix is adding value to their service by respecting their customer's time.
I've found movies through the current rating system that I liked but would never have picked out on my own. Conversely, I've watched movies in the theater that I hated and found Netflix's system reflected my disdain when they became available.
The ability to quickly and accurately sift through decades of media is a major value to this consumer.
(reply to this comment) (link to this comment)
Re: Yay!
Unnoticed? I don't think so. Perhaps the word you were looking for is 'unmentioned'.
The value of good recommendations has been discussed at length on this blog, hardly unnoticed.
http://www.techdirt.com/articles/20081123/1230162928.shtml
(reply to this comment) (link to this comment)
How does pandora do this? They seem to be very good at very similar task...
(reply to this comment) (link to this comment)
Re:
Pandora calculates a "musical DNA" for each song by scoring it on a multitude of criteria: beat, chorus, length, sound, instruments, etc.
Pandora then assumes if you like one song, you will like songs with similar DNA.
Netflix uses a more human approach called "Collaborative Filtering" in which they (generally) say "people like you who liked these films ALSO liked this film".
The main difference is Pandora analyses the music in detail, but Netflix analyzes people's opinions in aggregate.
(reply to this comment) (link to this comment)
I think what most impresses me here is that two teams stopped competing and decided to gang up on the problem, a problem solving attitude that is so rare these days.
(reply to this comment) (link to this comment)
a couple of points
There were at least three teams (not two) involved in the final result - BellKor, Chaos and Pragmatic Theory.
The winner has to tell the world how they did it but they are allowed to patent their technique if they can. The rules require disclosure and a non-exclusive licence to Netflix but they do not prohibit patenting and they do not prohibit charging others for the use of the technique.
(reply to this comment) (link to this comment)
I completely agree, this is a great way to foster innovation. Given the combined efforts of all of the teams $1 millon is a bargain.
(reply to this comment) (link to this comment)
I've heard that nobody finally have won this prize. And I've heard that two of three teems stopped competing. I have a research paper about this topic. If you want, I can check for you.
(reply to this comment) (link to this comment)
Add Your Comment