Design And Implementation Of A Chinese Natural Language Understanding Subsystem In A Car Question And Answer Field

In recent years,with the explosive development of artificial intelligence technology,more and more companies have begun to explore in the field of human-machine dialogue.At present,many companies have put dialogue robots into production,such as Baidu’s Xiaodu,Tmall’s Tmall Elf,Xiaomi’s Xiaoai classmates,Jingdong JIMI,etc.,which are used in smart home,robot customer service and other fields.In the natural language understanding(NLU)process of dialogue processing,most of the rules are used to extract the user’s intention and word slot.Although this method is simple and stable,as the rules increase,the cost of maintenance and expansion will increase dramatically.This paper proposes to use deep learning technology to design and implement a subsystem that provides training and management of Chinese natural language understanding model in the field of automotive question and answer,providing data management,model training,service configuration and NLU analysis services.Through the research on the related technologies of natural language understanding at home and abroad,and the related product experience and use in China,combined with the specific business scenarios,the requirements analysis of the natural language understanding subsystem was carried out,and the subsystems were divided into four modules,namely Training data management,model training,service configuration,and NLU resolution services.The training data management module includes two sub-modules of training corpus management,dictionary management and rule management,and provides input or deletion of training data for webpage visualization.The model training module includes two sub-modules,an intent recognition model training and an entity recognition model training,and an optional hyper-parameter used in the model training to start the business model training.The service configuration module displays all the trained intent recognition models and entity recognition models,selects the intent recognition model and the entity recognition model according to the requirements for natural language understanding analysis test,and starts the NLU service.The NLU Parsing Service Module is a Thrift service that receives invocation requests and provides parsing functionality.The system is implemented in Python,the model design uses the Pytorch framework,the remote service call uses the Thrift framework,and the Flask framework is used to build the web application.The natural language understanding subsystem shields the specific details of the model training,provides an easy-to-use interface for the model development user,and provides an efficient and accurate model for the service caller.