Translate

Tuesday, May 20, 2008

Netflix million dollar prize, an idea

I finally heard about a million dollar prize that Netflix is offering to improve its prediction of what movies people will like based on what they've said they like so Netflix can recommend movies to them. And they only want a 10% improvement on what their system does to get the million dollars.

Well, I'm not going to join, and I just started thinking about this a few minutes ago when I stumbled across it by reading another article, but it seems to me that most people don't have much of a clue about why they like something, but they are fairly good at what they don't like.

So just brainstorming and tossing something out there as I like doing such things and I'm not going to put in much effort here, I'd consider all the ways people do not like a movie, like it's too long, or the musical score is horrible, or it leans toward celebrities versus great actors, and figure out how to see dislikes in the recommendations.

Then I'd associate dislikes with things like subject, director, actor and use that to eliminate movies in a category that the person likes, which is the only place I'd go with what they've watched in the past.

So, like if they like drama, but their preferences show a dislike for the director Cameron, I'd have the system not suggest his movies, and I'd have it work opposite to what most people would do, and suggest the movies not eliminated by dislikes.

It just seems to me that dislikes can be more powerful than likes, as people avoid nasty things more than they seek things out.

And that's it for brainstorming on this issue.

If you think that idea can be tried and you want to do a lot of database heavy programming, feel free to take it! Go, have fun!

You can even keep all the money. I'd just want to be known as the source of the idea.


James Harris

No comments: