DIALOGUE STRATEGY LEARNING METHOD, PROGRAM, AND DEVICE, AND STORAGE MEDIUM

摘要

PPROBLEM TO BE SOLVED: To reduce the labor hour necessary for learning a dialogue strategy. PSOLUTION: The dialogue strategy learning device executes a procedure for accepting input of an evaluation value for a response to a user terminal in order to learn a Markov decision process; a procedure for integrating one or more similar states to one cluster paying attention to the frequency of transition of state; a procedure for forming a first Markov decision process from the evaluation value and the cluster by reinforcement learning; a procedure for decomposing the cluster of the first Markov decision process to the state before integration to form a second Markov decision process, and storing it in a storage means; and a procedure for outputting the second Markov decision process as a Markov decision process used in a dialogue system. PCOPYRIGHT: (C)2006,JPO&NCIPI