Using History of Belief States for Adaptive Dialogue Management
Xiaobu Yuan · 2019
This paper applies the theory of information space for an inquiry into dialogue management based upon the partially observable Markov decision process (POMDP) and proposes to maintain the history of belief states for the analysis of trend changes. The establishment of relationship between the number of changing trends in belief states and level of knowledge among users then directs a modification of POMDP-based dialogue manager to assist users achieving their goals by adaptively choosing different sets of policies according to the type of users. Results of experiment demonstrate that this modification allows embodied agents to reduce the length of dialogue while increasing the accuracy of intention discovery.