Capturing knowledge of user preferences with recommender systems

Capturing user preferences is a problematic task. Simply asking the users what they want is too intrusive and prone to error, yet monitoring behaviour unobtrusively and finding meaningful patterns is both difficult and computationally time consuming. Capturing accurate user preferences is, however, an essential task if the information systems of tomorrow are to respond dynamically to the changing needs of their users. This thesis tests the hypothesis that using an ontology to represent user profiles offers advantages over traditional profile representations in the context of recommender systems. A novel ontology-based approach to recommendation is applied to a real world problem and empirically evaluated. Synergy between recommender systems and ontologies is then explored to help overcome both the recommender system cold-start problem and the ontology interest-acquisition problem. Finally, the visualization of profiles in ontological terms is examined in a real world situation and empirically evaluated.