Design And Implementation Of A Natural Language Understanding Model Training System

With the development of artificial intelligence technology,dialogue system is more and more widely used.As an important branch of dialog system,task-based dialog system can accomplish many basic tasks and save human cost for enterprises.Natural language understanding(NLU)is an essential core module of Task-based dialogue system,which includes two key sub tasks: intention recognition and slot filling.NLU is usually implemented based on neural network model.In view of the wide application value of Task-based dialogue system and the indispensability of natural language understanding in Task-based dialogue system,a natural language understanding model training system is proposed to improve the development efficiency of natural language understanding model.The system is based on B/S structure and is used by a single NLU developer.Each developer deploys the system to their own development host,and can train NLU model through visual operation.First of all,through the demand analysis,the functions of the system are defined,and the overall design of the system is carried out.Then,the system is divided into four main modules: domain setting,data resource management,model training and model deployment testing,and the detailed design of each module is carried out,including UML class diagram design and functional flow chart design.The system server uses Python language,which is based on the flask framework and tensorflow framework.The front-end page is based on the vue.js framework.A variety of models are built in the system for developers to choose from,and the data enhancement function based on rules and dictionaries is provided.In addition,Pinyin features are introduced to alleviate the possible problems of speech recognition under voice interaction.The system modularizes the model training process,shields the specific code implementation through the web page,makes the work handover between NLU developers with different code styles more convenient,and greatly improves the model development efficiency of developers.