Research And Implementation Of Question Understanding And Answer Search In Insurance Intelligent Answering

Health insurance is a highly specialized,complex areas of knowledge,relevant knowledge for most people know little about the insurance field.Insured persons often have difficulties in comprehending the terminology,such as "the selection of insurance products" and "protection conditions".Intelligent answering technology can accurately understand the user's query intention and accurately locate the query results.It has become one of the important research directions in the field of artificial intelligence.This paper focuses on the research of the health insurance intelligent answering system,and analyzes the problems and implementation techniques of the health insurance answering system.Starting from the understanding of user query statements,a health insurance intelligent answering system is designed based on a structured health insurance knowledge base.At the same time,the paper make deep research on deep learning technology,and use related algorithms in the intelligent answering system to improve the accuracy of the system.The main research contents and contributions of this article are as follows:(1)Build insurance document collection and quality control moduleIn the process of collecting insurance clauses,there are a large number of problems such as "same names with the same name" and "more than one name".Using the string similarity matching algorithm and manual screening,the paper build an insurance clause acquisition and quality control module.To some extent,it avoids data redundancy and improves the quality of structured insurance data.(2)Design and implement an insurance intelligent answering moduleThe intelligent answering method for finding similar question answers cannot accurately understands the user's query intention and requires a large amount of manpower to construct the FAQ library.To solve this problem,this paper uses natural language technology to understand user intention and queries based on a structured insurance knowledge base.The the paper uses ElasticSearch and Django to build an insurance intelligent answering system.Firstly,the paper identifies the insurance name in the question based on the subject recognition model;Then the multi-pattern string matching algorithm is used to identify the relation entities in the question based on the relational synonyms dictionary.Next,The attribute entity in the question sentence is extracted by using separation and n-gram retrieval.Finally the question is converted into an ElasticSearch query,Query DSL,based on a custom rule template.The answer is searched from structured insurance knowledge.The system understands the user's query intention more accurately,saves the user's time for locating knowledge,and improves the user experience.(3)Research on subject recognition method of user's questionThe insurance name is too long,and there are a lot of abbreviations and ambiguities in the user's query.The insurance name cannot be identified by template and dictionary.Therefore,identifying the insurance naming information in the user's question becomes a research difficulty.In response to this problem,this paper use the subject recognition model(Bi-LSTM-CRF)combining conditional random field(CRF)and bidirectional long-term memory recurrent neural network base on the real health insurance domain question.And add pre-training words embedding to train this model.Then using the accuracy P,recall rate R,F1 value to evaluate the performance of the model.Experimental results show that the model of this paper obtains an accuracy of 82.3% in the identification of insurance subjects,which is improved compared to other models.Finally,this paper applies the Bi-LSTM-CRF model in the health intelligent answering system.