Netflix is running a contest to improve recommender prediction systems. A 10% improvement over their current system yields a $1M prize. Failing that, the best smaller improvement yields a smaller $50K prize. This contest looks quite real, and the $50K prize money is almost certainly achievable with a bit of thought. The contest also comes with a dataset which is apparently 2 orders of magnitude larger than any other public recommendation system datasets.
9 Replies to “$1M Netflix prediction contest”
Comments are closed.
I’m currently in Tom Mitchell’s ML class and of course this is announced just as it comes time to pick a semester project. Very tempting… Are recommender systems an interesting problem, aside from the money attached to them?
I consider them interesting. Recommender systems are an instance of vastly multitask learning, which is one direction that machine learning definitely needs to pursue.
It’d be interesting to list opinions on what people think Netflix’s motivation is. The following are meant to get discussion going (by no means the most important).
1. Genuine belief that better recommendations would pay back more than $1million (and that they’d still hold this belief in 2011 when the contest closed).
2. No belief that $1M would be worth it but $50k for marginal improvement is a steal ($50k is very small – less than one researcher’s salary when benefits are included).
3. Publicity (at low cost publicity if no one wins the $1M). Compare, for example, the free, large newspaper articles that have already appeared.
Others?
10% better performance would certainly be worth many millions to them.
I’m sure the publicity will earn them $1M back ! Not so sure if the improved recommendation system would make any difference…and in case no one wins it, so much publicity for 50k is almost free publicity !
I assume another motivation of Netfix is to let a number of researchers work for them with almost no salary. Imagine what it would cost for Netfix to employ all of us who participate in the competition in order to get the same results.
There is currently a poll ove on KDnuggets about when/whether this will be solved:
http://vote.sparklit.com/poll.spark/203792