Best Intentions? Towards Intention Recognition and Strategy Switching for Mixed-Initiative Recommender Systems

Recommender systems are designed to assist users in the search for product and service information and have been successfully deployed in a range of domains, from restaurants to route planning, movies to news. In particular, conversational recommender systems engage the user in an extended recommendation dialog, making suggestions and eliciting user feedback in order to guide the next round of recommendations. This complementary user-system interaction has led to a recent interest in mixed-initiative systems research. In this paper we examine some of the shortcomings of existing case-based conversational recommender systems. In particular, we highlight how a more flexible recommendation strategy, one that responds to intermediate recommendation success and failures, can lead to significant improvements in both the efficiency and quality of recommendation dialogs. We argue that such techniques have a role to play in mixed-initiative recommender systems in the future.