Spoken language dialogue strategy generation method and spoken language dialogue method

Abstract

The present invention provides a spoken language dialogue strategy generation method. The method comprises: the step S1) employing the mode of human being replacing a machine to collect a true man-machine dialog data sample; the step S2) based on an Agenda simulation user model, constructing a virtual user according to the dialog data sample for simulation of the true user behavior; the step S3) adding noise for the virtual user's semantic information, and constructing a noise channel; the step S4: constructing a dialogue strategy template according to the virtual user's semantic information; the step S5: extracting free parameters included in all the conditional statements in the dialogue strategy template to form parameter factors, and inputting the parameter factors into the genetic algorithm for optimization to obtain the optimal solution; and the step S6: applying the optimal solution to the dialogue strategy template to obtain a dialogue strategy. Compared to the current dialogue strategy made by hands, the spoken language dialogue strategy generation method has better noise robustness; and moreover, the dialogue strategy language defined in the invention is easy to artificial edition and maintenance and is more suitable for the business environment having strict requirement for the system behaviors.

Description

technical field [0001] The invention relates to the field of spoken language interaction, in particular to a method for generating a spoken dialogue strategy and a spoken dialogue method. Background technique [0002] With the rapid popularization of mobile Internet and smart terminals, spoken dialogue systems have been more widely and deeply used. Spoken language interaction is one of the most natural and convenient human-computer interaction methods, and is widely used in voice assistants, automatic customer service, automatic information query and other systems on mobile devices. Compared with traditional input methods such as keyboard and mouse, spoken dialogue systems have a lower threshold for use, a wider target user group, and are more suitable for human-computer interaction in special environments such as vehicles and wearables. Spoken dialogue management is the central module of the spoken dialogue system. Its main function is to receive semantic information from ...

AI Technical Summary

Problems solved by technology

At present, in commercial systems, domain experts mainly construct maintenance dialogue strategies manually, which is relatively cumbersome.

In the research field, models and methods based on reinforcement learning are often used to automatically optimize the strategy, which has better robustness; however, the trained strategy is a numerical model, which is difficult to manually adjust and verify. Less used in business environment

Images