Optimal Dialog in Consumer-Rating Systems using POMDP Framework

Abstract

Voice-Rate is an experimental dialog system through which a user can call to get product information. In this paper, we describe an optimal dialog management algorithm for Voice-Rate. Our algorithm uses a POMDP framework, which is probabilistic and captures uncertainty in speech recognition and user knowledge. We propose a novel method to learn a user knowledge model from a review database. Simulation results show that the POMDP system performs significantly better than a deterministic baseline system in terms of both dialog failure rate and dialog interaction time. To the best of our knowledge, our work is the first to show that a POMDP can be successfully used for disambiguation in a complex voice search domain like Voice-Rate.