On the Dynamic Generation of Compound Critiques in Conversational Recommender Systems

Conversational recommender systems help to guide users through a product-space towards a particular product that meets their specific requirements. During the course of a “conversation” with the user the recommender system will suggest certain products and use feedback from the user to refine future suggestions. Critiquing has proven to be a powerful and popular form of feedback. Critiques allow the user to express a preference over part of the feature-space; for example, in a vacation/travel recommender a user might indicate that they are looking for a “less expensive” vacation than the one suggested, thereby critiquing the price feature. Usually the set of critiques that the user can chose from is fixed as part of the basic recommender interface. In this paper we will propose a more dynamic critiquing approach where high-quality critiques are automatically generated during each recommendation cycle from the remaining product-cases. We show that these dynamic critiques can lead to more efficient recommendation performance by helping the user to more rapidly focus in on the right region of the product-space.