Design And Implementation Of Intelligent Air Conditioning Customer Service System

In recent years,the number of small and medium-sized enterprises in China has been increasing steadily year by year.Meanwhile,the customer service market has also shown a growth trend.The market scale of call centers and online customer service has exceeded 100 billion yuan.At the same time,with the development of the mobile Internet,people have more and more shopping channels,and the demand for customer service has also increased.Companies will suffer a decline in turnover and loss of customers if high-quality customer service cannot be provided.Therefore,how to provide efficient customer service that meets their own development needs at a lower cost has become a key concern for many companies.Traditional customer service is mostly manual,but manual customer service has problems such as high training costs,weak ability to serve multiple channels,and low efficiency of customer service personnel.The existing intelligent customer service mostly records frequently proposed questions in advance,and interacts by providing options for users to choose from.Such a way is quite different from the way people are adapted to dialogue and cannot solve the user's personalized needs.Therefore,it is of great significance to develop an intelligent customer service system that interacts in a natural language for easy customization and expansion.This thesis introduces the design and implementation of an intelligent customer service system in an air conditioner reporting scenario.The main function of the system is to identify the problem described by the user and provide a self-check method for operation,record the basic information of the air conditioner and the user,and determine the time for on-site maintenance with the user.The system mainly includes four modules.The first module is natural language understanding,using regular matching,sequence labeling,semantic analysis and other technologies to convert the natural language text input by the user into a triple of actions,slots,and slot values;the second module is dialogue state tracking,Update the current dialogue state according to the user's semantics;the third module is policy learning,the system decides the actions to be performed in the current round of dialogue through pre-defined rules or reinforcement learning;the fourth module is response generation,according to the actions to be performed by the system,natural language text is generated and returned to the user.The system uses the Python language for development,adopts the B/S architecture,and the development method of separation of front and back ends,which facilitates the maintenance and iteration of the system in the later period.The front-end uses Html5,Bootstrap and other technologies to implement user login and interaction interfaces.The back-end uses Flask as the Web framework,uses the relational database MySQL to store conversation records and basic user information,and uses Flask-SQLAlchemy as the ORM framework to combine the tables in the database with Entity objects in Python establish correspondence.The system uses Gunicorn to deploy to meet the needs of concurrent access by multiple users,and uses Jmeter for stress testing.In terms of algorithms,the system adds the BERT pre-training model to the natural language understanding module,and achieves a high recognition accuracy rate on the corpus that lacks annotations.