Supporting User Query Relaxation in a Recommender System

This paper presents a new technology for supporting flexible query management in recommender systems. It is aimed at guiding a user in refining her query when it fails to return any item. It allows the user to understand the culprit of the failure and to decide what is the best compromise to chose. The method uses the notion of hierarchical abstraction among a set of features, and tries to relax first the constraint on the feature with lowest abstraction, hence with the lightest revision of the original user needs. We have introduced this methodology in a travel recommender system as a query refinement tool used to pass the returned items by the query to a case-based ranking algorithm, before showing the query results to the user. We discuss the results of the empirical evaluation which shows that the method, even if incomplete, is powerful enough to assist the users most of the time.