Design And Implementation Of Task-based Dialogue System Development Platform

Task-based dialogue systems are widely used in various fields and products,which can effectively reduce labor costs while helping companies improve service quality and service experience.The traditional task-based dialogue system mainly has the following three problems:1)It is highly dependent on NLP experts.All links are scattered,and there is no engineering process.2)The dialogue management corpus is a strictly structured dialogue story stream that contains intention and slots and historical dialogue status information based on the original sentence input by the user.Manual labeling is difficult and it is difficult to guarantee the labeling quality.3)The traditional task-based dialogue system is difficult to flexibly reply questions by searching the knowledge base in the multi-round task-based dialogue process.In response to the above problems and challenges,this article engineered the process of building a task-based dialogue system.The main research contents are as follows:1)Design and implement a task-based dialogue system development platform.Decoupling business and technology,built-in AI capabilities to relieve expert dependence,and provide a clear and standardized engineering construction process.Developers only need to configure the intent,slot and tagging corpus to train a task-based dialogue system that supports multimedia responses and integrates QA knowledge base questions and answers by clicking the button once.2)Propose and implement a dialogue management plan and a method and tool for constructing dialogue management corpus.Design a tree-shaped dialogue flow structure to implement a dialogue story drag-and-drop editor,construct an executable dialogue story flow by arranging dialogue nodes,quickly and efficiently construct dialogue rules,and can be transformed into dialogue management corpus.In addition,design a dialogue management scheme that combines rules and deep learning models,so that the dialogue management part of the dialogue system can not only have the accuracy brought by the rules,but also have the learning ability of the deep learning model.3)Design a task-based dialogue system scheme that effectively integrates Q&A based on knowledge base into the multi-round dialogue pipeline structure,so that the dialogue system can efficiently and collaboratively process task-based multi-round dialogue and knowledge base-based retrieval question and answer.In addition,the QA question-and-answer module based on the BM25 preliminary screening module and the BERT-based fine ranking module is designed to ensure the accuracy of the system and effectively reduce the response delay.Finally,using this platform as a tool to build an intelligent customer service in the vaccine appointment scenario,it verifies the integrity,effectiveness and practicability of the platform and program in this article.