Jump-Starting MovieLens: User Benefits of Starting a Collaborative Filtering System with "Dead Date"
Date of Submission:
March 1, 1998
Collaborative filtering systems use a database of user preferences to recommend items to users. They suffer from a start-up problem: early users of the system enter preference data without receiving much value back. This paper reports on our experiences addressing the start-up problem in MovieLens-a movie recommendation web site created aspart of the GroupLens Reserch project. We were fortunate to have a database of over two million movie ratings available to seed our system, though these initial ratings could not be asociated with MovieLens users because of privacy concerns. We call such a collection of ratings from inactive users "dead data". This research looks at the effect of that data on the quality of the predictions made, and on user satisfaction. We are able to conclude that starting a collaboratve filtering system with "dead dadta" improves the user experience.