How to Learn More about Users from Implicit Observations¹

Abstract

In this paper, an approach to learning user interests is presented. It relies on positive evidences only, in consideration of the fact that users rarely supply the ratings needed by traditional learning algorithms, specifically not negative examples. Learning results are explicitly represented to account for the fact that in the area of user modeling explicit representations are known to be considerably more useful than purely implicit representations. A content-based recommendation approach is presented. The described framework has been extensively tested in an information system.