A belief function reasoning approach to web user profiling

Abstract

This paper presents a novel approach to web user profiling. Our proposed approach consists of two main parts. The first part focuses on discovering user interests in a user feedback collection, usually including relevant and irrelevant documents. Frequent pattern mining widely used in data mining community is applied to extract user feedback information. The second part is to represent user profiles. We introduce a novel user profile model based on belief function reasoning. In this model, the user profile is described by a probability distribution over the user feedback information extracted. Experimental results on an information filtering task show that the proposed approach clearly outperforms several baseline methods.