Posted by: Douglas Macmillan on September 21, 2009
On Monday, video rental service Netflix handed an oversized $1 million check to a team of software engineers who spent the last three years building an improved version of the company’s video recommendation tool. But the more valuable prize may be the sophisticated algorithm which Netflix now gets to implement on its site and which participant AT&T plans to adapt for its own uses.
The goal of the Netflix Prize, announced in 2006, was to improve the Web site’s ability to predict which movies customers will like by 10%. From a field of thousands of entrants, two teams completed this goal in June. The group calling itself BellKor’s Pragmatic Chaos — a joint effort by several of the top competitors — was declared the winner, since it was the first to achieve a 10.06% improvement in movie predicting.
Netflix founder and CEO Reed Hastings believes better predictions will help his company retain subscribers, at a time when viewing choices are expanding to include online upstarts like Hulu and Google’s YouTube. Currently, Hastings says most subscribers “really love” one out of three movies they watch; he expects the improvements created by the Netflix Prize winners will help the company raise that number to two out of three movies. “The more people love movies, the more they watch, and the more they watch, the more they love Netflix and stay with Netflix,” Hastings says.
Here’s a video interview I took of Hastings at the Monday morning event:
AT&T Research fielded its own Netflix Prize team, who eventually became part of the willing BellKor team. The company was less motivated by the million-dollar purse than by its own business interests, says Chris Volinsky, director of AT&T’s statistics research department. He says the data Netflix provided to entrants made the perfect test lab for AT&T’s own video recommendation service, which aims to let cable subscribers find channels and programs suited to them. “When we first started, the digital TV service was kind of a fledgling product — we didn’t have a lot of data. So being able to work on a real data set with real customers on movies was invaluable to us for developing our models at AT&T.”
Here’s my interview with Volinsky taken moments after he received his prize:
Netflix took the occasion to announce a second Netflix Prize, which will incorporate more of the company’s data, like the demographics of its users. The teams who have made the most progress in the new contest by April 2010 and by April 2011 will each be awarded with $500,000.
Will other companies take a cue from AT&T, and use the engineering competition to build their own recommendation engines?