BellKor’s Pragmatic Chaos says it can predict better than Netflix Inc. what ratings Netflix members will give movies.
Two-and-a-half years after challenging anyone to beat by 10% the accuracy of its system that guesses how members will rate movies they’ve not yet seen, Netflix Inc. has announced a team known as BellKor’s Pragmatic Chaos has laid claim to the $1 million prize. Per the rules of the contest, a ticking clock was started in early July that gave other teams 30 days to try and beat BellKor’s Pragmatic Chaos’ improvement factor of 10.05%.
The Netflix Prize seeks to substantially improve the accuracy of predictions on how much members are going to like or dislike a movie they have yet to see, based on their ratings of movies they’ve already seen. Netflix members rate movies on a scale of one to five stars. The more movies they rate, the easier it becomes for Netflix to guess how members will rate movies they have yet to see.
The BellKor’s Pragmatic Chaos team has submitted evidence that the predicted ratings its algorithms generate are 10.05% more accurate than those generated by Netflix’s current ratings prediction system, called Cinematch. If no other team matches or exceeds 10.05% before the 30-day period is up, then Netflix will initiate a validation process it says will take a few weeks. If its system is validated, BellKor’s Pragmatic Chaos will be awarded the $1 million prize.
BellKor’s Pragmatic Chaos is actually a team of teams. Four teams that had been working separately joined forces under the BellKor’s Pragmatic Chaos moniker. Team members include:
- Bob Bell and Chris Volinsky, members of the statistics research department at AT&T; Research.
- Andreas Toscher and Michael Jahrer, machine learning researchers and founders of Commendo Research and Consulting in Austria.
- Electrical engineer Martin Piotte and software engineer Martin Chabbert of Montreal.
- Yehuda Koren, senior research scientist at Yahoo Research.