urn:lj:livejournal.com:atom1:lastparagraph lastparagraph lastparagraph 2006-10-07T15:29:03Z urn:lj:livejournal.com:atom1:lastparagraph:852 2006-10-07T15:29:03Z 2006-10-07T15:29:03Z Netflix, the DVD rent-through-the-mail service, is hosting a <a href="http://www.netflixprize.com">competition</a> to improve their movie suggestion system, with a prize of .... $1,000,000 &lt;/dr.evil&gt;.&nbsp; Netflix provides around 100,000,000 historical ratings from their customers (masked, of course), with which to build a scoring algorithm.&nbsp; To claim the prize, the algorithm must predict another qualifying data set with an <a href="http://en.wikipedia.org/wiki/Root_Mean_Squared_Error">RMSE</a> of less than 0.8563, which is a 10% improvement over Netflix's current RMSE of 0.9474.<br /><br />Last night, having downloaded and briefly looked at the data, I submitted a very simple solution, which at the time <a href="http://www.netflixprize.com/leaderboard">placed me 10th</a> with a RMSE of 1.0004.&nbsp; I know I can do much better than that, given sufficient time to investigate &amp; build a better algorithm - of course I'm sure others can too.&nbsp; In the end, it's not really about the prize for me (though it would be nice), but rather the competition - seeing how well I can do relative to others.