Construction Of Knowledge Graph In The Insurance Field

With the improvement of the domestic quality of life,more people have begun to pay attention to various types of insurance to obtain further protection for their belongings and health.And with the rapid development of the Internet and the rise of artificial intelligence,product design and risk estimation in the insurance industry have begun to gradually move towards intelligence.Because insurance is closely related to human activities,and the diversity of human activities nowadays,various data and documents in the insurance industry are becoming more and more complex and diverse.Knowledge Graph is a technology that associates different data and knowledge together,because it can be more in line with the real real world,and it can broaden the dimensions of traditional data storage and business presentation methods,so from the initial search field A concept of concern has slowly become one of the technologies that are competing for research in all walks of life.By constructing a knowledge map in the insurance field,it can help insurance companies to comprehensively analyze customer information and achieve further optimization of products and investments through more effective risk assessment;it can also provide customers with more personalized and customized insurance products,and build more reasonable Insurance clauses,etc.So as to achieve a win-win situation for enterprises and customers.This article focuses on some key foundations of knowledge graph construction,and constructs knowledge graph for the insurance field,and establishes a knowledge system that can be traced.The focus is on the named entity recognition model based on deep learning to obtain relevant entities of the insurance industry from natural texts,establish the connection between insurance entities and natural texts,and realize knowledge traceability.The main work of this subject includes the following aspects:(1)Data collection and preprocessing.This part realizes the control function of multiple crawlers by realizing the crawler front-end control system and combining the distributed capabilities of the Scrapy framework.In addition,it has implemented website crawlers for many Internet websites such as news websites and insurance industry related websites,and collected hundreds of thousands of Internet web pages.Further analysis of the web page structure of this website has been done to extract the main text and key fields of the web page.Finally,the subject clustering and word frequency analysis are done on the main text of the webpage,which realizes the classification and extraction of insurance subject text,and lays the foundation for further entity extraction and other related work.(2)Implement a named entity recognition model based on active learning and the BERT model.This part first explores the BERT model,understands the structure and advantages of the BERT model,and proposes improvements to the pre-training of the BERT model for Chinese,and finally uses BERT The model builds a named entity recognition model.Then the query strategy of active learning is understood and analyzed.Aiming at the large amount of Internet data,a batch-type active learning training strategy is proposed and the effectiveness of this method is proved through experiments.(3)Realize the construction and display system of the knowledge graph,improve the existing knowledge graph construction method and combine with the characteristics of the insurance industry,by simplifying the construction steps,realizing the quick iterative update of the knowledge graph of the insurance industry,making it compatible with the Internet The characteristics of the era data.Then understand the characteristics of the Janu Graph graph database,and realize the storage and knowledge traceability functions of the insurance industry knowledge graph.Finally,functions such as dynamic display and online editing of the insurance industry knowledge graph are realized.