Improving the Performance of Recommender Systems That Use Critiquing

Personalization actions that tailor the Web experience to a particular user are an integral component of recommender systems. Here, product knowledge – either hand-coded or “mined” – is used to guide users through the often overwhelming task of locating products they will like. Providing such intelligent user assistance and performing tasks on the user's behalf requires an understanding of their goals and preferences. As such, user feedback plays a critical role in the sense that it helps steer the search towards a “good” recommendation. Ideally, the system should be capable of effectively interpreting the feedback the user provides, and subsequently responding by presenting them with a “better” set of recommendations. In this paper we investigate a form of feedback known as critiquing. Although a large number of recommenders are well suited to this form of feedback, we argue that on its own it can lead to inefficient recommendation dialogs. As a solution we propose a novel recommendation technique that has the ability to dramatically improve the utility of critiquing.