Title:
REINFORCEMENT LEARNING AGENT FOR MULTIDIMENSIONAL INTERACTION ACTIVITY SELECTION
Document Type and Number:
Japanese Patent JP2023074463
Kind Code:
A
Abstract:
To provide a method and a device that generate a response within an interaction system.SOLUTION: A method includes: a step for receiving the vocalization of a user; a step for updating an interaction state on the basis of the vocalization; a step for generating information to identify the interaction state; a step for, by a first interaction activity agent, selecting a first activity candidate from a first set of activity candidates using a first machine learning model and the information to identify the interaction state; a step for, using a second interaction activity agent, selecting a second activity candidate from a second set of the activity candidates using the second machine learning model and the information to identify the interaction state; a step for, by an evaluation agent, selecting an output activity using a third machine learning model, the first activity candidate, and the second activity candidate; and a step for generating a system response on the basis of the output activity.SELECTED DRAWING: Figure 5
Inventors:
SIMON KEIZER
SVETLANA STOYANCHEV
RAMA SANAND DODDIPATLA
SVETLANA STOYANCHEV
RAMA SANAND DODDIPATLA
Application Number:
JP2022136157A
Publication Date:
May 29, 2023
Filing Date:
August 29, 2022
Export Citation:
Assignee:
TOSHIBA CORP
International Classes:
G06N20/00
Attorney, Agent or Firm:
Patent Attorney Corporation Suzue Patent General Office
Previous Patent: Communication terminal and communication system
Next Patent: Release layer-forming composition and method for producing release film
Next Patent: Release layer-forming composition and method for producing release film