Abstract
Conversational recommender systems have been introduced in Travel and Tourism applications in order to support interactive dialogues which assist users in acquiring their goals, e.g., travel planning in a dynamic packaging system. Notwithstanding this increased interactivity, these systems employ an interaction strategy that is specified a priori (at design time) and is followed quite rigidly during the interaction. In this paper we illustrate a new type of conversational recommender system which uses Reinforcement Learning techniques in order to autonomously learn an adaptive interaction strategy. After a successful validation in an off-line experiment (with simulated users), the approach is now applied within an online recommender system which is supported by the Austrian Tourism portal (Austria.info). In this paper, we present the methodology behind the adaptive conversational recommender system and a summarization of the most important issues which have been addressed in order to validate the approach in an online context with real users.