14. October 2007

MovieLens: Personalized Movie Recommendations with Stunning Accuracy


Screenshot of MovieLens

How do you know whether you will like a movie or not? You could read reviews or ask friends – but unfortunately movie critics as well as friends usually have a different taste in movies than you.

MovieLens wants to help at this point. MovieLens is an internet project of the university of Minnesota. It can predict for almost every movie how much you personally are going to like it. With this basic feature, it provides a set of other functions like recommendation lists or group recommendations.

To make such accurate predictions, MovieLens must first get to know your taste in movies. You have to rate as many movies as possible in the systems by giving them 0.5 to 5 stars. It is recommended to rate at least 35 movies, but the more you rate the more accurate the predictions will be.

The trick behind MovieLens is called “collaborative filtering”. Basically, it uses its vast database of user rating to find users which have a similar taste. If these users like a movie, chances are good that you will like it as well. Being a research project, the real trick is slightly more complicated, but this is the basic idea (as far as I know). The same basic idea is used in various music recommendation systems.

It is even possible to get recommendations for whole groups – provided that each group member is also a MovieLens member. This can ease the planning of the next video night. In group mode, MovieLens gives a combined prediction, but the predictions for each single member is also displayed.

The accuracy of these recommendations is stunning. I rated more than 100 movies. Based on that data MovieLens can usually predict my rating for a movie with an accuracy of 0.5 stars.

It also provides a nice search function which allows you to only receive recommendations in certain genres from specific production years and more.

In short: Cool stuff, with the downside that you have to invest some rating work before you can use it effectively.

» MovieLens

(this article is also available in German)

2 Comments »

  1. It started with Amazon. Once I found I could get recommendations based on my interests (loosely… see Amazon specific review below) I became quite the “movie rater”, rating what I had seen just to see what else would come up. Then I found much more devoted websites, my favorite of which is MovieLens because of the community. Now I’ve been with several rating sites for quite a while, and they’ve all had time to delve into my interests.

    Comment by Which Movie To Watch Next ? — 26. October 2007 @ 9:05

  2. But,

    How do you really know that the results are what you wanted. A rating average must be made and applied.

    John

    Comment by John Ortega — 17. July 2009 @ 9:31

TrackBack URI

Sorry, the comment form is closed at this time.

Powered by WordPress

Subscribe to RSS Feed
blogoscoop