14 DAY TRIAL // The Netflix Prize has finally come to an end. Last month one team managed to enter a recommendations algorithm that was 10 percent better than Netflix's and claim the $1 million prize. Since then all teams were allowed 30 days to enter better scores and make their claim at the prize. Yesterday entries were closed, with two teams duking it out till the last minute.
Last month a team made up of members of several top groups, BellKor’s Pragmatic Chaos, entered a score that was 10.08 percent better than Netflix's recommendations system, moving the contest to the final phase. The 10.08 percent score stood until the 25th, when a team called The Ensemble entered a score that was 10.09 percent better, overtaking BellKor, less than a day before the competition would close. Team BellKor wasn't done yet, though, and managed to enter a score that tied The Ensemble's with 24 minutes left. And just when everything seemed over The Ensemble entered a new score, which was 10.10 percent better, again overtaking BellKor, within 4 minutes of the contest's closure.
“We are delighted to report that, after almost three years and more than 43,000 entries from over 5,100 teams in over 185 countries, the Netflix Prize Contest stopped accepting entries on 2009-07-26 18:42:37 UTC. The closing of the contest is in accordance with the Rules – thirty (30) days after a submitted prediction set achieved the Grand Prize qualifying RMSE on the quiz subset,“ read the final announcement, posted soon after The Ensemle's final entry, on the site set up for the competition.
But things aren't done yet as the public score is only for the so-called “Quiz” test, with another set of data, called the "Test," available only to Netflix. The final winner of the competition would be determined based on both scores and won't be announced until Netflix gives out the prize within a few weeks. Yehuda Koren of BellKor Pragmatic Chaos apparently posted that his team had the lowest Test score though this hasn't been confirmed by Netflix.
The Netflix Prize was set up in October 2006 challenging the world's leading computer scientists to come up with a recommendations algorithm that was at least 10 percent better than the one the company used internally. It sparked the interest of numerous experts around the world and progress was fast at first. But, as the teams closed in on the 10 percent mark, improvements came much slower and in smaller increments. In January this year the top score was already at 9.63 percent yet it took six months to get past 10 percent.