Exploiting learning techniques for the acquisition of user stereotypes and communities

In this paper we examine the acquisition of user stereotypes and communities automatically from users’ data. Stereotypes are built using supervised learning (C4.5) on personal data extracted from a set of questionnaires answered by the users of a news filtering system. Particular emphasis is given to the characteristic features of the task of learning stereotypes and, in this context, the new notion of community stereotype is introduced. On the other hand, the communities are built using unsupervised learning (COBWEB) on data containing users’ interests on the news categories covered by the news filtering system. Our main concern is whether meaningful communities can be constructed and for this purpose we specify a metric to decide on the representative news categories for each community. The encouraging results presented in this paper, suggest that established machine learning methods can be particularly useful for the acquisition of stereotypes and communities.