Design And Implementation Of Insurance Question Answering System Based On Deep Learning

With the rapid development of the Internet and the explosive growth of information,the intelligent question answering system,which obtains the information needed by users from the massive information also emerged.Users can ask questions through the question answering system to get simple and accurate answers.Different from search engines that require users to search and integrate relevant information to get answers,the question answering system can get an accurate and short answer,able to meet the needs of quickly getting answers.Question answering system has been widely used in practice,especially in the limited field that can provide professional and accurate answers to the questions raised by users in the actual social production.Most of the initial question answering systems were based on rules and statistical models,which could not have a good understanding of natural language questions.With the development of technology,deep learning,an important branch of the machine learning field,has emerged.It uses neural networks as feature extractors to free people from traditional machine learning heavy feature engineering.Deep learning has achieved good results in the field of question answering systems.Most of the intelligent question answering systems with the best performance are based on deep learning.This paper constructs an insurance question answering system based on deep learning,which aims to provide a professional,efficient and accurate question answering system for users who want to learn knowledge about insurance.The system includes three modules:insurance Q&A dataset construction,candidate Q&A pair retrieval,and answer generation.The insurance Q&A dataset construction module is responsible for the collection,expansion of the dataset,and the construction of the Q&A index database.The candidate Q&A pair retrieval module is responsible for understanding the questions raised by users and quickly retrieving candidate Q&A pairs based on Lucene from the built Q&A index database.The answer generation module is responsible for the generation of answers.The candidate Q&A pair set is further screened by the BERT model to obtain a smaller high-quality candidate answer set.The BertSum model is used to generate a text summary of the candidate answer set to produce the final answer.The system adopts a three-layer architecture design.The front end is built with the Thymeleaf template and Bootstrap framework,which is responsible for user-interactive operations,sending requests to the back end and presenting the returned data.The back end is divided into Java middle-stage and Python back-stage.The Java middle-stage is implemented with the SpringBoot framework and Mybatis framework,which is responsible for the information interaction with the front end and back-stage,index library construction,and the implementation of data persistence.The Python back-stage is responsible for the code implementation of deep learning.gRPC is used for cross-language communication between the middle-stage and back-stage.Use the MySQL database to store system data.