The New York Times reports today that Netflix is holding a contest to improve their recommendation engine. Anyone who can improve the service’s recommendation function by a mysterious 10% will win a $1 million prize. One easy way to start would be to look at MovieTally, a tag-based movie recommendation site built by a 15 year old from New York named Hayden (he asked that his last name not be used).
MovieTally lets users add their own tags and reviews to movies found in the user generated movie database. It’s easy to click around and find movies that other users have tagged with the same terms as your favorites. The site recommends users who have similar favorites lists to your own. I’ve been looking at this site for awhile it was really sparse before getting some good coverage at OpenGardens for its implementation. Now that it’s has begun to gain traction, the database has been bulked up and the service is much more useful.
Netflix currently lets users view the movies selected by friends, but as a poster child for moving services to the web I’d love to see them take up the most current social technologies. Tagging allows for exploration by theme in a much more flexible way than formal site created categories. It would be good if instead of asking for improvements to its black box recommendation engine Netflix began by making movie discovery more transparently social.
See also our review of BiggerBoat, a media search service for site publishers that presents loads of metadata. It’s a different approach that you’re liable to start seeing implemented at sites around the web, as is the case at allmovie.com already. There are many ways that Netflix could improve its already fantastic service but newer startups provide easy examples.
It seems Netflix could have an interest in not owning a proprietary movie recommendation engine, but instead to support an open one and just be able to use the results. That way they aren’t limited to just what their users think.
Are there any details on this contest? I don’t really understand how it would work. Anybody can come up with a theory on how to make it better but without actual testing how do you determine. Is Netflix just going to let people test their theories directly on the site?
Nice! I’ve been looking for a good tagging one for a while. I think the key thing for movies is going to be when you’re able to tag which friend you want recommending which type of movies. Some friends are better than others for different genres. (Have I missed it? Is there one that does this yet?) This to me will beat an algorithmic one any day (although I give Netflix credit for trying since I like the service).
If all this does is get Netflix a bunch of publicity (which it is), it was more than worth the $1m they may not ever give out.
That should be “allmovie.com” (singular). Allmovies.com (plural) goes to some kind of spam/aggregator website.
in response to Patrick: they are giving out “annonymized data”. In light of the AOL search debacle, this may not seem too smart, but you have to register for the contest before you get access to the data, and they don’t rent porn, so I can’t imagine that anyone would be too embarrased.
there are more details on the site.
It seems that Blabbit.com has some of their answers already. I agree with Michael about checking out the web to see what’s out there.
The problem I see with using websites such as MovieTally for recommendations, is that in the end you tend to end up mirroring the box-office hits: what most people watch, most people recommend.
So, unless there’s an additional user-controlled component, such as “show me only recommendations from these group of users,” you’ll only get junk.
A “circles-of-trust” algorithm, where users give a particular user’s recommendation more (or less) weight would work best. I have very good friends who’s recommendations I don’t follow at all. Some other friends, I know I can trust for certain genres only. Some mere acquaintances give wonderful movie suggestions. Now, if I could call on them while at the video store… that would be great. I’m sure Netflix could pull this off.
I think this is much more of a marketing effort than a serious attempt to find better recommendation logic.
The website for more information is http://www.netflixprize.com
This is clearly an inexpensive mktg ploy by Netflix or a disillusional attempt to garner developer support for a service that is past it’s prime.
While competitors, upstarts and the likes of YouTube et al compete for eyeballs and mktshare, Netflix is still in the ‘bricks & morter’ mode of video distribution. Unless they get their act together fast, they will likely fade into oblivion as more efficient means of video consumption gain critical mass.
nemrut…I’d love to hear the rational behind anything you just said. Because it’s a bunch of baloney. Granted, this may be some marketing ploy. But if you’re implying streaming video rentals are going to take over standard rentals, at any point in the near future. Then you’ve lost it.
_70% of Americans are online and growing
_TV is no longer the sole medium for passive video consumption
_Streaming video quality via Movielink, Cinemanow, YouTude, etc is not great, but good enough
These are a few reasons i can think of for Netflix’ demise if they don’t get with the program. Granted they won’t go out of business anytime soon, but their current busines model is on a path to incrmental if non-existent growth.
With the popularity of the svcs mentioned, people want video on-demand now and without a lot of fuss. This is made even more convenient with the popularity of laptops surpassing that of desktops.
I used to think tht Netflix was the next best thing to apple pie, but with the inconvenience of physical delivery, returns, late,misplaced,wrong disks it’s just not worth the time and effort. with so much quality video on the net competing for eyeballs along with limited time and bandwidth, who wants to deal with the hassles of ‘physical delivery/fulfillment’…
movietally seems pretty trivial – it’s yet another collaborative filtering system with tags in addition to simple ratings. highly doubt it’s anywhere near as sophisticated as netflix’s system. not every problem can be solved with tags, sometimes you need good old fashioned real computer science. the average netflix user probably wouldn’t bother with tags anyway – it’s far simpler to rate 1-5, so given that that’s the bulk of the data netflix has, they want to maximize the efficacy of their recommendation system that uses that data. the contest is a great idea, and will attract the right minds. no, it’s not web 2.0, no there’s no nifty ajax, but it has the potential to contribute important results to further research. this is an artificial intelligence problem, and web 2.0 junkies need to get over themselves – grouping and matching with tags is too shallow to do anything special.
Very good move on the part of Netflix to engage the community…
It time to move beyond the various extensions of collaborative filtering mechanisms. I would imagine that the way to improve upon the existing recommendations is by breaking the movies down along a lot richer set of attributes (similar to the Music Genome project http://pandora.com/mgp.shtml). Based on better understanding of movies, better recommendations should be possible.
For something that made Amazon what it is now (the recommendation engine) I think a million dollar is pretty cheap.
It will be interesting to see what brilliant ideas will come out of this.
nemrut: “Not great but good enough”? Sorry, I would still prefer to watch a movie on the tool that was designed to show video content… TV (preferable my plasma) I’ve neither the tiime nor the patience to wait for video download. It take me hardly any time to order a movie from Netflix. Not sure about you but my time is worth more to me than to go through the hassle of downloading video.
Mark Cuban has written about this with considerably more insight than those that have no experence in this medium but think they do.
http://www.blog...ick-biz-lesson/
Here’s the money shot from Cuban’s post:
“And as far as the perspective of giving users what they want and they want downloads. If we dont offer them downloads, they will find downloads on their own , just as we saw in the music business… Nope. This aint the music business. This aint 3 minute songs that download in under a minute and allow users the option of getting the 1 song they like instead of a package of 10. IF movies were sold in prepackaged albums of 10 movies. Maybe. If movies were 3 minutes in length. Maybe. If watching on a computer or on an Ipod was as good an experience and often better than watching on the smallest TV in the house, maybe. But its not.
Watching video on a computer or on a PDA/Ipod is a 2nd class experience. It works amazingly well as a time killer on a bus, plane, lunchroom. It works good enough in a dorm room or your apartment bedroom, but its not going to replace watching on a real tv. It will always be a niche market in every manner.
and for all you “but i can and do download everything ” types. Good for you. Get up and away from your computer and go see how the rest of the country lives. And when you hit 27, get a real job or move out of your parents house, whichever comes first, tell me if you are still downloading 10 movies and burning them for your friends and creating 10 playlists for Itunes every week like you did in college or when you first graduated. At some point you realize the time you spend downloading and burning to a DVD is worth more than the 10 bucks to go to a movie or 20 bucks or less to buy the movie. When that happens you will have figured out that all that time you spend burning DVDs and trying to manage space on your hard drive wasnt worth it”
The bummer is that there is no user attributes. Simply the age and gender would probably improve recommendations greatly.
Though I think it’s a great opportunity for research, but I doubt that it will lead to any major changes in its current implementation. User rating data is insufficient for building an intelligent recommendation system.
http://harry.hc.../2006/10/03/391
What do you guys think? – HC
ABCota – You saved me from writing my own response this morning. Thank You.
Netflix does throttle and often I have notice dvds are supposed to be shipped one day and then it is changed to the next. Customer service is not great at all, but the software for selecting movies is the best I have seen. I tried Blockbuster on line and it was okay, but selection software was not as good and due to location, a day was added each way for receiving and returning movies. (Their free instore rental did me no good as the nearest stoe was 23 miles away) INTELLIFLIX WAS THE WORST. They double charged me and kept $177 of my money for days. I cancelled the account and ended up be charged over $40 for 3 movie rentals even though the crooks had illeaglly kept $177 of my money for days when they had no way of knowing if I might have needed that money for rent, a car payment, etc. It could have even caused checks to bounce. Fortunately it did not, but it could have. BEWARE OF INTELLIFLIX. Do not try it as a alternative to Netflix!
Netflix has its problems, but it is the best of the three online dvd rentals I’ve tried.