When Netflix, the Web-based DVD-rental company, offered $1 million to anyone who could significantly improve its predictions about how much a customer will like a movie, the contest generated buzz among computer scientists. Money was part of the draw, recent graduate David Weiss ’07 admits, but the theoretical and technological challenges were also alluring. Equaling Netflix’s existing system seemed difficult. Improving its accuracy by 10 percent — the threshold for the top prize — would be monumental.

Weiss found two friends who were eager to give it a try: Pyne Prize winner Lester Mackey ’07 and Weiss’ roommate, David Lin ’07, a mathematics major. The three downloaded 100 million lines of test data and began working in October 2006. By graduation, they had climbed to the top 50 on a leaderboard that included more than 2,000 entrants.

The Princeton team, named “Dinosaur Planet” for the last movie listed in the test code, was an underdog against top competitors that included a pair of Ph.D.s from AT&T Research and an academic lab at the University of Toronto. But with a voracious appetite for research and ample free time during the summer, the Princetonians made a steady climb to compete for the 2007 “progress prize,” a $50,000 award for the best team as of Oct. 1.

Late in the competition, the Princeton trio joined forces with a group of Hungarian graduate students and jumped to first place a day before the Oct. 1 deadline, using algorithms that improved on Netflix’s accuracy by more than 8 percent. But the AT&T Research duo recaptured the lead on the final day and won the progress prize.