LEARNING SYSTEM, SIMULATION DEVICE, AND LEARNING METHOD

摘要

PROBLEM TO BE SOLVED: To solve the problem that an interaction device interacting with a user needs a large amount of labor to construct a weight vector required to output a sentence.

SOLUTION: In this learning method, interaction sentence information is received from the interaction device, a method identifier and interaction probability information belonging to the interaction sentence information are used to determine a user sentence type, one piece of information or more in one or more spots or one or more determination factors belonging to the interaction sentence information are used to acquire the one or more determination factors or the one or more spots, a user sentence type identifier and user input information having the one piece of information or more in the one or more determination factors or the one or more spots are transmitted to the interaction device, a reward calculated by using a coincidence degree between a user preference vector and one or more evaluation values of the spots included in the user input information is used to update a weight vector corresponding to the method identifier of the interaction device. The learning method can automatically construct a weight vector needed for the interaction device interacting with the user to output a sentence.

COPYRIGHT: (C)2012,JPO&INPIT