A Million Dollar Contest - For Super-Geeks

Netflix is a very large (largest?) DVD rental company. One of their most important assets is Cinematch, an in-house developed movie recommendation system. Cinematch's purpose is to predict, based on a user's previous rating of movies, the rating a user would give to other movies. The result, of course, is a recommendation of movies that are most likely to suite the user's taste.

Apparently, Cinematch works pretty well, but the guys at Netflix would like it to work even better. So they came up with the Netflix Prize: encourage developers and researchers to come up with an algorithm that improves the quality of rating prediction of Cinematch significantly enough (at least by 10%), by promising a prize of 1,000,000$. To make it even more interesting, and since the whole contest spans over at least 5 years, there is a yearly prize of 50,000$ which will be given to the best solution each year.

Being a Machine Learning freak, I find this contest GREAT! I also like the rules of the contest a lot. Basically, the winning algorithm will have to be made publicly available at the end of the contest.

This kind of initiatives are so great because they encourage the development of new concepts and algorithms, which even if they won't win the first prize, might very well be helpful for various other purposes. I think the best comparison is Fermat's Last Theorem. In his will, Paul Wolfskehl initiated a prize of then 100,000 marks to whomever would be able to prove or disprove Fermat's Last Theorem. This generated a huge interest in the subject, which resulted in an incredibly rich amount of new ideas and whole new areas of mathematics being discovered and researched to this day. I doubt that the Netflix Prize will have the same effect, but I do believe it will give Machine Learning a well-deserved boost.

As far as Netflix is concerned - they can only win from it. The news about this contest should inevitably increase their exposure. If the contest succeeds, and someone manages to provide significantly better results - it will be worth much more than 1M$ for them. If nobody manages to win the contest, then they can heartily claim to be using the best movies matching algorithm human brain could come up with do date. Either way, it's a win-win situation for them.

It's a very difficult task, but I think I'm going to give it a try, as far as time permits...

If you're interested, following are the Terms and Conditions in a Nutshell:

Contest begins October 2, 2006 and continues through at least October 2, 2011.

Contest is open to anyone, anywhere (except certain countries listed below).

You have to register to enter.

Once you register and agree to these Rules, you’ll have access to the Contest training data and qualifying test sets.

To qualify for the $1,000,000 Grand Prize, the accuracy of your submitted predictions on the qualifying set must be at least 10% better than the accuracy Cinematch can achieve on the same training data set at the start of the Contest.

To qualify for a year’s $50,000 Progress Prize the accuracy of any of your submitted predictions that year must be less than or equal to the accuracy value established by the judges the preceding year.

To win and take home either prize, your qualifying submissions must have the largest accuracy improvement verified by the Contest judges, you must share your method with (and non-exclusively license it to) Netflix, and you must describe to the world how you did it and why it works.

For more elaborated information, check out the Netflix Prize page.