Data mining support for case-based collaborative recommendation

This paper describes ongoing research which aims to enhance collaborative recommendation techniques in the context of PTV, an applied recommender system for the TV listings domain. We have developed a case-based perspective on PTV's collaborative recommendation component, viewing the sparsity problem in collaborative filtering as one of updating and maintaining similarity knowledge for case-based systems. Our approach applies data mining techniques to extract relationships between program items that can be used to address the sparsity/maintenance problem, as well as employing recommendation ranking that combines user similarities and item similarities to deliver more effective recommendation orderings.