Finding optimal presentation sequences for a conversational recommender system

Abstract

This paper presents an approach for finding optimal presentation sequences in conversational Recommender Systems. The strategies simultaneously pursuit the goals of acquainting the user with the different possibilities, successfully accomplishing the task in the shortest possible time, and obtaining an accurate user model. The approach is modeled as an MDP where the states include belief states about the acceptability of the different alternatives, modeled as Bayesian networks. © 2012 Springer-Verlag.