Design And Implementation Of Automotive Intelligent Customer Service System

Intelligent question answering system is an artificial intelligence product that deeply integrates natural language technology and information retrieval technology.As a special field,the automotive field uses big data,cloud computing,artificial intelligence and other cutting-edge technologies to innovate service and management models,and has become a new direction for the development of automotive related industries.The system obtains the data published on the Internet through crawlers,and then transforms the RDF data to construct a knowledge relationship graph.At the same time,deep learning combined with dictionary matching model is used in the core part of question answering system such as entity recognition,entity link and user intention recognition.In the implementation of entity recognition and intent recognition algorithms,the use of BERT word vectors for sentence-level information utilization improves the recognition accuracy.Compared with the traditional Word2 Vec method,the recognition accuracy is improved by about 6 percentage points under the current corpus.On this basis,the system adds a dialogue state management mechanism to improve the accuracy of response on a certain basis.Finally,after the module and function test of the system,the system showed the response accuracy and load performance in line with expectations.The main research work of this article is as follows:1)Knowledge base constructionThrough the crawler to obtain car data on many websites on the Internet,and then build ontologies based on knowledge research in the automotive field,then perform data cleaning,data filling,knowledge fusion,etc.to complete the construction of the knowledge base.2)The core algorithm design of the question and answer system in the automotive fieldObtain data through car-related problems,and construct a data set in combination with manual annotation.The system uses BERT training word vectors to improve the representation ability of word vectors and improve the division of word boundaries;in order to solve the problem of uneven distribution of the number of effective entities,an attention mechanism is also used to improve the classic named entity recognition model;In the process,standardized entities are used to identify the user’s intentions,the user’s expected actions are obtained,and the answers to the questions are queried from the knowledge graph by combining the entities and the user’s intentions.3)Implementation of question answering systemThe system can provide Web services and interface services to meet the needs of different types of users.The system can achieve second-level response for a single user,and can support 50 people to use online at the same time,which can better meet the user’s needs for system performance.