[The following is an exact transcript of this podcast.]
Netflix isn’t satisfied with the way its system recommends new movies to customers based on their viewing habits. So the mail-order DVD rental company has offered outside teams prizes to improve its accuracy. A group from AT&T Laboratories has already won $50,000 for figuring out a formula that’s 8.43 percent better at telling a film buff what to rent. And Netflix is sweetening the pot—the team that can improve recommendation accuracy by 10 percent will get a cool million.
The contest requires that recommendations be made using the ratings customers give other movies they’ve rented. But the researchers say whether or not a person explicitly rates their returns, their rental history can be used as an “inferred rating” of things like genres or actors. What’s more, the preferences of other customers can predict how someone with similar rental histories would score a film. The research is explained in the May issue of the journal IEEE Spectrum.
There are certainly bigger problems to solve these days than recommending movies. But it would be nice to know why Netflix keeps insisting after I’ve returned Slumdog Millionaire and Delicatessen that I’d really like Annie.