October 2nd, 2006
Netflix, the easy-to-use mail-in DVD service, is offering a 1 million dollar prize to anyone who can create a better movie recommendation system than their current one. According to the NYTimes, Netflix will offer the prize to anyone who can cull through their gigantic data set and come up with a system that improves the current version by at least 10%.
That’s a tough job, given that the Netflix web site is nearly a pure-play recommendation system, meaning that without the recommendations feature the site is only a shell of its former self. The recommendations system is what drives Netflix. Roughly 2/3 of all rented movies there come from recommendations.
As far as building a better system, Reed Hastings, Netflix’s CEO, admits that they’ve hit a wall: “If we knew how to do it, we’d have already done it…And we’re pretty darn good at it right now. We’ve been doing it a long time.”
To bootstrap the contest Netflix is making a huge part of its ratings database public so contestants can deal with real data and results can be judged objectively. Though they’ve taken major precautions with their data, this is a daring move after the recent AOL debacle, wherein AOL made part of their search queries database public and even casual browsers could easily figure out who the queries belonged to. That shouldn’t happen in this case, though, and even if it did the data should not be as sensitive as AOL’s.
I think this is a really good move by Netflix. They’ll get a little press out of it as well as a better recommendation system and a better service, benefitting them two-fold. I wonder what other, similar ways companies could do something like this, open-sourcing innovation?
This could have broad effects over many industries. For many of us, recommendations are how we find out about and decide to try something new (not just movies, music, and books). We might get a restaurant recommendation from a friend, or a digital camera recommendation from a geeky cousin. We do this to save ourselves time…it would be impossible to do good research on all the items we’re interested in. Recommendations are a shortcut to good information, and most of the time are well-considered. I’m really interested in them not just because I’m speaking about them at UI11, but because I think we’ll start to see a much broader adoption throughout all sorts of web applications.
However, I do wonder how one might go about improving on Netflix’s system. One way would be to have better social data. For example, right now I only have a couple Netflix “friends” in the system, simply because I haven’t bothered to ask the people I know who use it to link up. If those people are who I listen to when it comes to recommendations, then their presence as a friend in the system should definitely improve my recommendations because it better models my current habits. However, this data cannot be part of the database offered by Netflix because it would instantly identify who it belongs to. That’s a paradox of social web sites: there’s an inverse relationship between their ability to recommend things to you and the amount of information you provide. As quality recommendations go up, privacy goes down.
At any rate, Netflix is running the contest for one year. Starting today.Tweet