Integrating User Behavior and Collaborative Methods in Recommender Systems

For recommender systems to be successful, they need to achieve a certain level of accuracy in their recommendations that is acceptable to the users. In order to achieve higher levels of accuracy, several researchers advocated the integration of the collaborative and the content-based filtering approaches [Balabanovic & Shoham 1997, Konstan et al. 1998, Pazzani 1999]. In fact, Pazzani [1999] shows that the system that combines the two approaches achieves 71% accuracy of predictions vs. 61% for a pure content-based and vs. 57% to 69% for a pure collaborative approach (depending on parameters used in the collaborative filtering approach).