|Mathematician Crunches Big Numbers of Netflix Fans to Predict Movie Preferences
February 27, 2009
by Emily Matras ’12
Netflix, the online movie rental giant, serves an impressive 8 million customers a month. However, this number one online DVD rental service wants to leave customers more satisfied. To do so, Netflix is sponsoring a “One Million Dollar Challenge” to improve its movie rating system. Davidson College Assistant Professor of Mathematics Tim Chartier hopes to contribute a solution.
While Chartier is not yet formally entered in the challenge, he is applying the prestigious Alfred P. Sloan Fellowship he received last year to research on the issue. The fellowship provides $50,000 to early-career researchers for two years of study in math and science.
|(l-r) Chartier and Kreutzer puzzle through their research problems.|
“We really weren’t originally interested in answering their questions, we just wanted to use the Netflix data,” said Chartier. “Only recently have our questions started to merge with those of Netflix. It’s hard to find data sets as large as theirs, so their challenge is valuable to me pedagogically.”
Chartier is adapting sports ranking algorithms, like those used to predict bowl games, to movie data on 17,000 users that Netflix provides to those interested in the challenge. The data includes what ranking users have given movies, and when the movies were ranked. “Our current direction is two-fold,” said Chartier. “We are working to apply the sports ranking algorithms to movies, and we are ranking users rather than movies. This should enable us to find users who are ‘like you,’ and better predict what movies you like.”
Netflix currently uses the Cinematch rating system to suggest movies to a user based on how much he or she liked other movies. Netflix wants to move toward a new system in which movies are suggested to a user based on ratings given by similar users. To win the $1 million, a program must be better than Cinematch at predicting what movies a user will enjoy, and must beat Cinematch’s prediction by at least 10 percent.
Chartier’s approach is to rank movies by putting them in competition with one another, like sports teams. “When one team beats another, rankings shift. Not only do the two teams that played each other move in the rankings, but so do other teams that have previously played them,” explained Chartier. “To apply that to movies, we had to have them compete. That’s where users’ ratings come in. The ‘score’ is what rating a user gives a movie, and the movie with a higher rating ‘wins.’”
Chartier is working on the project with a colleague at the College of Charleston, and has also employed Erich Kreutzer ’10 as a research assistant. He also plans to hire a graduate assistant to help this summer.
Kreutzer explained that Netflix doesn’t want to use a rating system purely based on either users or movies. “Netflix doesn’t necessarily want to know the top 100 films over the entire user populations,” said Kreutzer. “They want to know what films a specific user would enjoy.”
Kreutzer continued, “In a big picture sense, Netflix wants to replace the movie rental clerk in a retail store. A good clerk would use his or her knowledge of films to suggest films you may enjoy. In making good suggestions, the clerk increases the likelihood you will return to the store.”
Chartier and Kreutzer meet often to review their progress and to brainstorm ideas. Kreutzer’s role is to write the code to implement the ideas he and Chartier discuss, and evaluate results and report them to Chartier. “Dr. Chartier has many ideas on where research should go, but he also thoughtfully considers my ideas,” Kreutzer said.
Chartier says that they are still a few weeks away from running data to compare results with that of Cinematch. “It takes a long time to create and run algorithms because we can’t use something standard that tells us what to do. We’re creating something new mathematically,” he said.
But the long, laborious process is what Chartier loves about his research. “We have to create an algorithm, code it, and run it, and then it usually doesn’t work the first time,” said Chartier. “But that’s my favorite part, figuring out why a system doesn’t work. There is both an art and a science to it.”
Chartier’s research in computational mathematics and ranking systems applies to other areas as well. He is currently working on information retrieval and how the search engine Google ranks Web pages. Chartier is also applying sports ranking algorithms to their original purpose. “We’re actually going to use some methods we’ve been researching to predict results of this year’s NCAA basketball tournament,” he said. “Then we’ll submit them to the ESPN Challenge. I don’t really care if we win—this is more of an academic thing for me—but it would be cool if we did!”
|Chartier received an Alfred P. Sloan Fellowship to support his research.|
According to Director of Grants and Contracts Mary Muchane, Chartier’s research in numerical analysis is not often pursued outside of large research university settings. Muchane said, “In many ways, Professsor Chartier is a very rare bird. Computational mathematicians typically are not found at liberal arts colleges. The fact that he has active collaborations with scientists at several research universities and national laboratories strengthened his position considerably in applying for the Sloan Fellowship.”
Davidson is a highly selective independent liberal arts college for 1,700 students located 20 minutes north of Charlotte in Davidson, N.C. Since its establishment in 1837 by Presbyterians, the college has graduated 23 Rhodes Scholars and is consistently regarded as one of the top liberal arts colleges in the country. Through The Davidson Trust, the college became the first liberal arts institution in the nation to replace loans with grants in all financial aid packages, giving all students the opportunity to graduate debt-free. Davidson competes in NCAA athletics at the Division I level, and a longstanding Honor Code is central to student life at the college.
Posted By: Bill Giduz