Almost three years ago, in October 2006, Netflix, the online movie-rental company, announced a contest challenging the world's leading computer scientists to come up with an algorithm for recommendations that was better than the internal one by at least ten percent and advertising a $1-million prize. One team has just posted a solution that improves the recommendations by 10.5 percent.

As the contest began, progress was rapid, but, as the teams approached the ten-percent mark, they slowed down considerably. In fact, in January 2009, the leading team was at 9.63 percent and it took another six months to pass the mark. The team BellKor's Pragmatic Chaos is made up of members of the four other leading teams, with researchers from AT&T, Yahoo! Research Israel, Commendo Research and Consulting in Austria and Montreal's Pragmatic Theory. They will have to wait for a “buffer” period of 30 days now to see if another team can pass them in this time, after which the prize is theirs.

Netflix launched the contest because its recommendation system was one of the most important features of the site. In fact, 60 percent of the company's rentals come from its current system, called Cinematch. The system is especially useful for older or smaller titles that wouldn't otherwise be known to its users. In fact, the “long tail” movies, the less known titles rented by a small number of people compared with the site's user base, make up 70 percent of the company's rentals.

The company provided the contestants with huge amounts of anonymous user data and, based on ratings the users give to movies, the algorithm has to predict what they would also like. It took the combining of the top four teams to make it trough the ten-percent mark, and now they'll just have to wait and see if their efforts can be bested.