Task Oriented Dialogue System

With the continuous development of artificial intelligence technology.Dialogue system has been paid more and more attention by people because of its strong applicability and extensive application scene,and gradually stepped into every aspect of our life,speech recognition and synthesis,natural language processing,and so on.The development of science and technology,such as machine learning,deep neural network and so on,also accelerates this process,and makes the machine closer to the goal of smooth dialogue with people.Generally speaking,the realization process of dialogue system includes speech recognition,natural language understanding,dialogue management,dialogue generation and speech synthesis.Dialogue management is the core control of dialogue system,which guarantees the smooth progress of man-machine dialogue,and the knowledge base plays an important role in the accuracy of task oriented dialogue in the limited field.This paper mainly completed the following work.(1)Cross-platform cloud service model.This paper designs and implements a multi-terminal cross-platform dialogue system based on cloud server mode.The cross-platform cloud service enables users to use system functions on multiple terminals and realize data sharing,while the cloud server is flexible.The easy-to-expand features make the system easier to maintain.From the design point of view of system engineering,this paper determines the dialog function in the weather field of the system.The modular management of Xiaoai chat robot is based on the framework of the dialogue system.(2)Dialogue management design.In this paper,the framework-based dialogue management method is used to establish a task schedule for dialogue,and the process of obtaining task information in the process of human-computer interaction is regarded as a process of filling out the table.In order to ensure the accuracy of the state records in the dialogue process,the accuracy of the completed information should also be checked in the dialogue management process.The dialogue management module will also formulate dialogue management strategies from different design objects,such as the definition and verification of slots,the order of dialogue clarification,the reuse of historical information,and so on.(3)Knowledge base design.This paper analyzes the necessary knowledge content in the selected task field,and obtains the basic data of the knowledge base by collecting the network corpus extensively and selecting and extracting the data by hand.Then the representation method of template matching and frame representation is used to structure the knowledge base data to facilitate the reasoning and application of knowledge.