Using Collaborative Filtering Data in Case-Based Recommendation

In the context of PTV, an applied recommender system operating in the TV listings domain, we are examining the potential benefits in merging case-based and collaborative filtering (CF) recommendation techniques by developing case-based reasoning (CBR) methods that employ collaborative filtering style ratings profiles directly as cases. Doing so presents a number of challenges, both in applying a case-based perspective to collaborative filtering, and in addressing the sparsity problem that plagues many collaborative filtering systems. This paper expands on earlier CBR views of collaborative filtering, identifies problems and opportunities for similarity maintenance therein, and proposes and evaluates methods for mining and applying new similarity knowledge.