Improving Recommendation Effectiveness: Adapting a Dialogue Strategy in Online Travel Planning

Abstract

Conversational recommender systems support a structured human-computer interaction in order to assist online tourists in important online activities such as travel planning and dynamic packaging. In this paper we describe the effects and advantages of a novel recommendation methodology based on Machine Learning techniques. It allows conversational systems to autonomously improve an initial strategy in order to learn a new one that is more effective and efficient. We applied and tested our approach within a prototype of an online travel recommender system in collaboration with the Austrian Tourism portal (Austria.info). In this paper, we present the features of this technology and the results of the online evaluation. We show that the learned strategy adapts its actions to the served users, and deviates from a rigid initial strategy. More importantly, we show that the optimal strategy is able to assist online tourists in acquiring their goals more efficiently than the initial strategy. It can be used by the system designer to understand the limitations of an existing interaction design and guide him in the adoption of a new one that is capable to improve customer relationship, the usage of their web site, and the conversion rate of their online users.