Million Dollar Netflix
“We’re quite curious, really,” says NetflixPrize.com. “To the tune of one million dollars.” Netflix has announced a competition and promised to award the winner $1 million, but the question it seeks to answer is hardly the type found on most game shows. The internet DVD service, facing the challenges of competitors like Blockbuster and looking ahead to the onset of online movie downloading, is seeking to improve the movie recommendation service. Its million dollar challenge goes out to a team that can do just that.
Its current program, “Cinematch,” bases its recommendation on a user’s history, looking both at the movies ordered from Netflix and also the ratings the user affixed to them after viewing. It then creates a list of titles from Netflix’s catalogue of over 65,000 movies that a given user would be likely to enjoy. Netflix’s 5.2 million members contribute about 2 million ratings per day, always feeding more information into the formulae and algorithms that fuel Cinematch’s mathematical predictions.
To sign up for the competition, interested parties must register at NetflixPrize.com, the site that outlines the rules for the Prize. Registered teams receive the Contest training data and a qualifying test sets of 100 million anonymous movie ratings. By 11 AM this morning, over 600 teams of 1,000 people from 31 different countries had registered to participate.
In order to win the $1 million Prize, the accuracy of the predictions submitted via a new method must be at least 10% better than the current Cinematch recommendations. In the event that no one meets this standard within the first year of the competition, a $50,000 Progress Prize will be awarded to the individual or team who improved the recommendation accuracy by the greatest amount. The Progess Prize will continue to be awarded until the 10% goal is reached, with the winner of the previous year’s Progress Prize setting the standard for the next year, should the competition run on for decades to come.
A chief stipulation in order to win either prize is a willingness to share not only the final product but also the way in which it works with the general public. Netflix has no plans to monopolize the new recommendation process; winning also requires that you license the method to Netflix, but not via exclusive licensure.
Netflix also guaranteed the anonymity of all of the reviews, in order to allay the privacy fears of any of its users. Text reviews have also been eliminated as possibly distinguishing factors, leaving only the titles, star ratings, and dates.
Netflix is turning innovation over to the public in part because those who engineer Cinematch are stymied; the method witnessed a period of rapid advancement which has since plateaued. A new eye (or 1,000 new pairs of eyes) to the process is precisely what Netflix desires. “Recommendation systems covering a wide variety of categories will play an increasingly significant commercial role in the future,” said Reed Hastings, the Co- Founder, Chairman and CEO of Netflix. “Right now, we’re driving the Model T version of what is possible. We want to build a Ferrari and establishing the Netflix Prize is a first step.”
Alexander McCabe has an interesting assessment of the current method’s accuracy and his opinion of the competition’s future; read about it at his blog, Flaunt It.
October 4th, 2006 at 12:15 pm
I didn’t care much for Netflix’s Rolling Roadshow movie festival as a promotional event, but this is a solid publicity stunt on the cheap and it certainly has generated a lot of media attention.