Collaborative Filtering
Arnd Kohrs

This project is a joint cooperation with ECHO, Sophia-Antipolis, France.

Project description

Collaborative Filtering is a technique by which the interest of a user for an object is predicted from the knowledge of the interest of other users for this object. It is a basic technique to build Recommender Systems on the Internet, where people can advise other on good and bad informations. One major problem of Collaborative Filtering is that a large database of user's advices is needed for a reliable prediction. We are investigating techniques to make a robust prediction even in cases where little data is available. This is of great help in the initial phase where a system is setup, and when a new user joins the system. As an illustrative example we are building a Active WebMuseum, a personalized navigation within a collection of painting, where the system can advise you for painting that you are likely to appreciate, based on the ratings from previous visitors.