johnkemeny.com

      

A couple of months ago Marcus Hutter offered up to 50,000 euros for losslessly compressing the 100MB English wikipedia corpus better than the then 18MB record.  Since then there has been a 6.8% improvement, and counting – each 1% is worth 500 euros.

Now Netflix is offering $1,000,000 for the first person or team to publish an algorithm that can predict how a customer will rate a movie based on prior ratings 10% more accurately than their current algorithm.  Be warned, the training files are big. There have already been 20 valid prediciton files submitted in the last 24 hours, and over 6000 teams from 92 countries registered.  So hurry!

The movie with the most ratings: Batman Begins.

Update:  Within 1 week the Netflix algorithm (subject of 15 years of research) has been bested.

Update2:  Some have claimed that the problem gets exponentially harder as you drive toward 10%.  We are almost half way there in under 3 weeks!  If linear we should be done by Thansgiving – but even if exponential, we can forecast victory in about 3 months.

This entry was posted on Thursday, October 5th, 2006 at 12:07 pm and is filed under Math & Science, Movies & Television. You can follow any responses to this entry through the RSS 2.0 feed. Both comments and pings are currently closed.