2 Comments

Summary:

(Updated to include the previously overlooked “The Ensemble”) Netflix posted an announcement yesterday that it has closed its Netflix Prize recommendation improvement contest. Though no official winner has been announced, the two leading teams, The Ensemble and BellKor’s Pragmatic Chaos (Team Pragmatic Theory and Team Bellkor […]

netflix_logo(Updated to include the previously overlooked “The Ensemble”) Netflix posted an announcement yesterday that it has closed its Netflix Prize recommendation improvement contest. Though no official winner has been announced, the two leading teams, The Ensemble and BellKor’s Pragmatic Chaos (Team Pragmatic Theory and Team Bellkor in Chaos joined forces) have supposedly cracked the Netflix nut by improving recommendations by 10.5 percent. You may yawn, say to yourself “Whatever, 10.5 percent, recommendations, blah blah blah.” But recommendations will play an increasing part of your viewing experience in the coming years.

To give you a sense of the size of this endeavor (and how difficult it was to find a winner), according to a post on the Official Netflix Blog, over the past three years there have been 44,014 entries from 5,169 teams in 186 countries vying for the top prize. The New York Times magazine ran an excellent feature last year about the challenges faced by the teams in the competition.

For Netflix, improved recommendations means you’ll rent more movies, be happier with Netflix and ideally keep the company on its hot stock streak. But this will go beyond Netflix.

The company has been pretty open with the whole process, and appears to be open about sharing the results as well. From the Netflix Prize FAQ:

Why the non-exclusive license?

First, we want to verify for everyone that the code did what was claimed; that means looking at it. And then we want to use it if we can. We’re a business and we want to make sure we can capitalize on the discovery. But we don’t want to impede the winner’s ability to capitalize on it as well. Actually, we hope they can build their own business and license it to others as well. That is the point after all.

That’s good news. You, as someone who presumably enjoys watching video, will increasingly rely on recommendations in the future. Thanks to the Internet, we’re moving from having 500 or so channels at our disposal to an almost infinite number of them. As I wrote in my recent long view post for our subscription service, GigaOM Pro, the goal in creating a workable user interface for televisions will be to hide most video options available in favor of showing you just what you like — or what you might like.

Now that this initial hurdle has been overcome (and the winners are $1 million richer) (well, a fraction of a million dollars, depending on how many people get that payout), hopefully recommendation innovation will keep up with the brave new overwhelming world of video.

  1. Chris,

    The problem with the winning solution(s) is that they are not practical and will not be used in real life.
    The conclusion from the Netflix prize is that no team was able to develop one “super algorithm” that improves Netflix’s by 10%.
    Eventually many teams had to join forces and combine as many algorithms as possible to create a clumsy solution that will not be practical in real-time environments.
    I believe that the real solution will come from companies like ours – BeeTV (www.bee.tv) or our competitors, companies that are focused not only on a mathematical answer but are researching and developing solutions that are looking at the whole picture of pushing content (not only movies!) to humans.

    Share
  2. @yaniv

    Nonsense, just because they used ensemble approaches doesn’t make the algorithms impractical. Ensembles have been seen to do well in many machine learning problems in the last decade.

    The leaders submitted multiple times in the last hours of the contest, which suggests that their algorithms run in a few hours, at most. How often does Netscape have to recompute recommendations for a subscriber, every day, perhaps? This seems eminently possible.

    Share

Comments have been disabled for this post