Design And Implementation Of Domain Task-Oriented Dialogue System

Since the advent of computers,how to better perform human-computer interaction has always been a topic of concern.Dialogue systems allow machines to communicate with humans using natural language like humans.Task-based dialogue systems are designed to help users complete specific tasks.The dialogue system constructed by the traditional method is difficult to have good performance in scenarios with scarce data sets,cannot adapt to changes in dialogue scenarios,and cannot answer common questions.The task-based dialogue system has attracted more and more attention because of its ability to reduce the consumption of manpower.Therefore,the design and research of the field-oriented task-based dialogue system is of great significance.The goal of this article is to design and implement a task-based dialogue system suitable for vertical fields.Focusing on this goal,this article conducts research on the algorithm and software architecture of the task-based dialogue system,and proposes new methods and solutions for the shortcomings of the existing technology.The main contributions of this article are as follows:1.Task-based dialogue system algorithm(1)In order to allow the system to have the ability to answer common simple questions while conducting multiple rounds of dialogue,design an algorithm framework that integrates single round of dialogue and multiple rounds of dialogue.This method firstly judges whether the dialogue is a single round of dialogue or multiple rounds of dialogue through intent matching,and then uses different algorithms to conduct the dialogue.Simulation experiments show that the fusion of a single round of dialogue effectively improves the task completion rate of the dialogue system and can respond to common simple questions.(2)Aiming at the inherent data hunger problem of deep learning methods,a dialogue rule tree data structure is proposed to represent multiple rounds of dialogue rules,and a dialogue management algorithm is established on this basis.At the beginning of multiple rounds of dialogue,this method locates the dialogue rule tree through the user's intentions identified by natural language understanding,and locates the nodes in the tree by the identified slot values,and then conducts the dialogue nodes through the identified slot values during the dialogue.The transfer guides the user to complete the task.Simulation experiments show that under the condition of limited annotation corpus,a higher task completion rate than deep learning methods is obtained.2.Microservice Architecture SystemImprove the concurrency and availability of the system,adopt the aggregator micro-service model to design the dialogue system,call the natural language understanding service,dialogue management service and single-round dialogue service through the dialogue agent service to realize the dialogue function,and design a service degradation strategy to ensure that the system is at a high level.The core dialogue function is available in the case of concurrency.The cache is constructed through Redis,and the data is pre-read from My SQL to the cache,avoiding the high latency problem caused by frequent access to the disk.The performance of the dialogue system was verified in the final functional test and performance test.