Can a machine know what movies you like?

Posted July 8th, 2009 at 11:51 am by Prabhakar Raghavan, Yahoo! Labs

 3 Comments / Filed in: Trends & News, Working at Yahoo!

If you’ve seen “The Godfather,” chances are you might like other Marlon Brando movies. Or films about gangsters. Or those directed by Francis Ford Coppola. But will you like “Napoleon Dynamite”?

This is the central problem posed by the Netflix Prize. Netflix is offering $1 million in prize money to anyone who can substantially improve (by more than 10 percent) the accuracy of its movie recommendation engine. While Netflix suggests movies based on your ratings history, the company isn’t satisfied with how well it can predict what you’ll like.

At Yahoo! Labs, this is just the kind of crazy difficult problem we love to take on. For scientists, it’s a pure challenge, requiring deep study and experimentation across a variety of fields, such as machine learning and data mining.

And for Yahoo! as a whole, these types of scientific problems also happen to be a critical element of what we most want to succeed at: connecting you with the content and information you most want in your life – even if you don’t know it yet.

That’s why we couldn’t be happier to pass along the news that Yehuda Koren, one of our scientists at Yahoo!’s Israel Lab, is part of the first qualifying team for the Netflix Prize.

Yehuda’s team, BellKor’s Pragmatic Chaos, reached first place on the Netflix Prize leaderboard on June 26, with an improvement of 10.05 percent. Achieving a more than ten percent improvement in the quality of movie recommendations is no drop in the bucket. It took Yehuda and his teammates three years to achieve and no other team has matched it yet.

Congratulations to Yehuda and his team. In the past few weeks alone, in addition to the Netflix Prize, Yehuda and his colleagues also received best paper prizes at two of the most important scientific conferences (ACM SIGMOD and ACM SIGKDD) for computer science and the Internet. Yahoo! researchers Christopher Olston, Shubham Chopra, Utkarsh Srivastava, Ashwin Machanavajjhala and Bee-Chung Chen, were also recognized for contributions to the science of how to better query and mine data, which will ultimately make it easier for you to get things done on the Web and beyond.

We may not yet have solved every problem the Internet has thrown our way, but at the very least, you should start feeling a lot more confident about those movies in your Netflix queue.

Prabhakar Raghavan
Head of Yahoo! Labs

Tagged: yahoo research, yahoo! labs

Post a Comment Bookmark This Digg This

3 Comments Add your own

Katherine | July 13th, 2009 at 5:05 pm

The Netflix is considerably getting better the exactness of predictions about how much somebody is going to love a movie, and Netflix is doing great so far, with the help of yahoo team one can at least consider that yahoo can achieve what they are up to.

John T Hawley | July 14th, 2009 at 1:33 pm

I am really amazed that yahoo is undertaking such an
enormous project.You people could not solve the simple
problems that I asked to have solved.I guess they were
not as important.If many people experience these
difficulties you might have to find ways to satisfy
some disgruntled customers if you plan on staying in
business.I have been a long time customer.I am beginning
to wonder why.

John T Hawley

gag | July 15th, 2009 at 6:51 am

this is tough, one of the many solutions

lets say a user watched the movie Troy

and lets say the recommendation engine has millions of movies to choose from then each movie behaves like a app or an element with points of characteristics in them like
genere, timestamp, popularity, ratings and the over all points
ex:
movie Click
genere- comedy – 50
timestamp- recent- 40
popularity- 20
rartings- 40
overall- 150

now all movies/apps have a lifecycle or their presence in front of movie troy which is random. Now if click happens to come up it is then matched against troy against various characteristics and overall ratings and based on the percentage score the movie is recommended.