Netflix’s 2006 Contest To Improve Movie Suggestions Became Informally Known As?
Answer: The Napoleon Dynamite Problem
In 2006, Netflix dangled a sizeable purse in front of freelance programmers the world over: significantly improve Netflix’s suggestion algorithm and win a cool million dollars. Seems like an easy enough proposition, right? After all, Netflix has a great algorithm but with a million dollars on the line it certainly should have been a walk in the park for the teams that raced to snag the prize.
What individuals and teams around the world quickly found was that the more they improved their algorithms, the slower the return on investment and the slower they inched toward the prize-winning goal of a 10 percent improvement. The community that had sprung up around the race to improve Netflix began calling this problem “The Napoleon Dynamite Problem” after the quirky 2004 indy comedy.
The movie proved to be so polarizing that predicting whether or not a Netflix user would like it was nearly impossible. No matter how similar the user profiles of two given users were (and thus how close their enjoyment of Napoleon Dynamite should have been), the movie proved to be, over and over again, an absolute wild card. The ratings on Netflix certainly reflect that status. Of the millions of times Napoleon Dynamite has been ranked by Netflix users, the ranking is almost always either one star or five. It’s a film you either delight in or absolutely despise.
What’s ironic is that by 2009 when a group finally won the prize, it didn’t matter as much to Netflix as they had originally anticipated. When Netflix set up the contest back in 2006, they were looking for ways to improve the suggestion algorithm for an audience that was renting DVDs. By 2009, things were shifting toward streaming and in the ensuing years Netflix found that the way people engaged with the media had changed. Watching a video through Netflix was no longer a multi-day affair for the majority of users; they were no longer debating over a movie, requesting it, waiting for it to arrive, and then watching it without any observation from Netflix. Instead they were selecting movies and shows on the fly and all the while Netflix could determine if they watched the whole thing, watched part of it, or bailed immediately.