Conversational recommenders with adaptive suggestions

We consider a conversational recommender system based on example-critiquing where some recommendations are suggestions aimed at stimulating preference expression to acquire an accurate preference model. User studies show that suggestions are particularly effective when they present additional opportunities to the user according to the look-ahead principle [32]. This paper proposes a strategy for producing suggestions that exploits prior knowledge of preference distributions and can adapt relative to users ’ reactions to the displayed examples. We evaluate the approach with simulations using data acquired by previous interactions with real users. In two different settings, we measured the effects of prior knowledge and adaptation strate-gies with satisfactory results. Categories and Subject Descriptors H.1.2 [Models and Principles]: User Machine Systems—human factors, software psycholog