Netflix takes cues from UC San Diego

One million dollars will go to the programmer who most improves the Netflix system for recommending movies based on how much a customer liked or disliked other movies.

Computer science and engineering professor Charles Elkan—a long-time organizer of UCSD data mining contests for students and researchers—helped write the Netflix Prize rules. Elkan suggested an online leaderboard as a way to keep the contest interesting. He also recommended a final month-long phase as soon as one contestant improves the accuracy of the current Netflix system by 10 percent, the minimum advance required for the grand prize: "If you want a competition to promote the growth of knowledge, you let people know it's feasible and see if anyone else can reach the goal." Elkan also pushed for a separate and confidential data set for determining the winner, in order to guard against algorithms geared specifically to the training data or leaderboard data, a phenomenon known as overfitting the data.