Abstract
Recommender systems are intelligent E-commerce applications that assist users in a decision-making process by offering personalized product recommendations during an interaction session. Quite recently, conversational approaches have been introduced in order to support more interactive recommendation sessions. Notwithstanding the increased interactivity offered by these approaches, the system employs an interaction strategy that is specified apriori (at design time) and followed quite rigidly during the interaction. In this paper, we present a new type of recommender system which is capable of learning autonomously an adaptive interaction strategy for assisting the users in acquiring their interaction goals. We view the recommendation process as a sequential decision problem and we model it as a Markov Decision Process (MDP). We learn a model of the user behavior, and use it to acquire the adaptive strategy using Reinforcement Learning (RL) techniques. In this context, the system learns the optimal strategy by observing the consequences of its actions on the users and also on the final outcome of the recommendation session. We apply our approach within an existing travel recommender system which uses a rigid, non-adaptive support strategy for advising a user in refining a query to a travel product catalogue. The initial results demonstrate the value of our approach and show that our system is able to improve the non-adaptive strategy in order to learn an optimal (adaptive) recommendation strategy.