Three years ago, Netflix announced that it would give $1 million to whoever could improve its recommendation algorithm, Cinematch, by 10 percent accuracy. Called the Netflix Prize, its difficulty had become infamous. Now, a team has finally accomplished the seemingly impossible feat--well, two teams actually.
After three years of competing, BellKor's Pragmatic Chaos submitted its winning algorithm only 24 minutes before another team, The Ensemble, submitted its own showing 10 percent improvement. In fact, the accuracy of the two submissions were identical, so BellKor was awarded the $1 million prize for being first. BellKor's Pragmatic Chaos is made up of seven members, including two AT&T researchers.
Given the success of this first competition, Netflix has set its sights on an even more difficult task for its sophomore effort. While the first competition focused on predicting which movies people would like from a set of sample user ratings, this one asks teams to predict based on demographic and behavioral data. Netflix says that it will provide teams with over 100 million data points, including age, gender, ZIP code, and previously rented movies. Unlike the original competition, this one is time-based, with $500,000 being awarded to the team who comes in first after six months and $500,000 being awarded to the team in the lead after 18 months, at the end of the contest.
From the looks of the new Netflix Prize competition, we should see some great advances in behavioral research. The task in this competition appears to be much harder than the one in the original competition, so it will be interesting to see what sort of progress the teams can make in 18 months.