Hacking NetFlix

Stanford Data Mining Class Students Working on Netflix Prize

Anand Rajaraman's students at Stanford are working on the Netflix Prize in his data mining class.

Different student teams in my class adopted different approaches to the problem, using both published algorithms and novel ideas. Of these, the results from two of the teams illustrate a broader point. Team A came up with a very sophisticated algorithm using the Netflix data. Team B used a very simple algorithm, but they added in additional data beyond the Netflix set: information about movie genres from the Internet Movie Database (IMDB). Guess which team did better?

Team B got much better results, close to the best results on the Netflix leaderboard!! I'm really happy for them, and they're going to tune their algorithm and take a crack at the grand prize. But the bigger point is, adding more, independent data usually beats out designing ever-better algorithms to analyze an existing data set. I'm often suprised that many people in the business, and even in academia, don't realize this.

Did your homework ever give you a shot at $1 million dollars?

Netflix Awards $50k Netflix Prize to KorBell Team

KorBell, a group of AT&T Bell Labs researchers, have won the $50,000 "progress" award in the Netflix Prize Contest (which means that the $1 million grand prize is still up for grabs).

The KorBell team of Yehuda Koren, Robert Bell, and Chris Volinsky, who work on visualizing and analyzing large networks for AT&T in Florham Park, N.J., improved upon the Netflix recommendation system by 8.43 percent. They led a crowded field of more than 27,000 contestants on more than 2,550 teams from 161 countries.

Koren, KorBell's captain, said he is an enthusiastic Netflix member who was drawn to compete for the Netflix Prize after the Netflix Prize dataset gave him an "opportunity to study a fascinating network." Koren said he and his team spent 2,000 hours in data analysis and computation to come up with the highest improvement to date over the Netflix recommendation system. Results of the team's work will also apply to AT&T Labs' ongoing research and development efforts to enhance the company's network and communications and entertainment service portfolio.

The press release has more details about the news.

Netflix Prize Approach Shared

Paul Harrison on Logarithmic.net has decided to share his approach to winning the $1 milion Netflix Prize, How to get an RMSE of 0.8937 in the NetFlix Challenge.

The current NetFlix leader has an RMSE of 0.8847, which my approach won't be able to beat no matter how much computing power I throw at it. There is therefore no longer a reason for me to keep my approach secret.

I did my calculations using Python and a small amount of inline C++, using the Numpy and Weave libraries.

My approach is based on simonfunk's SVD method. That is, we predict a user's rating of a movie by taking the dot product of two vectors, one corresponding to the movie and one corresponding to the user.

Are you working on the Netflix Prize? I'd love to hear your story.

Can A Philosopher Win the Netflix Prize?

Jeff Larche thinks a philosopher might be able to solve the Netflix Prize:

With an additional 3.87 percentage points yet to be racked up, the road to victory is long – possibly impassable. If I understand my statistical modeling correctly, every unit of progress to 10 percent will be a far tougher slog than the one before it. There clearly needs to be a breakthrough in how the problem is approached if anyone has a chance of winning. A couple of days ago, it occurred to me that the source of this breakthrough might be a better ontology.

Ontology is the study of logically structured categorical models. It helps us understand a particular domain of reality by looking at its essential elements, and especially, how they are interconnected. Because ontology proposes to explain big complicated things, this discipline was honed first by philosophers.

The Wikipedia has an article about Ontology.