The Architecture of Innovation

3

The Changing Face of Corporate R&D

IN 2006, NETFLIX, the video rental and streaming service provider, offered a prize of $1 million to the first team that could improve upon its movie recommendation algorithm by a set amount.¹ Up to that point, Netflix had relied on its Cinematch recommendation system to suggest movies. This system analyzed the accumulated ratings (on a five-star scale) of users to make personalized suggestions based on their tastes. Cinematch’s performance was rated by comparing the quality of the model’s predictions of a subscriber’s taste and the actual rating made by the viewer after she saw a movie (more technically, by computing the root mean squared error, or RMSE, of the predicted ratings). Over time, internal enhancements ...