Research On Vehicle Insurance Question Answering System Based On Deep Learning

In recent years,people's awareness of car insurance is gradually improving,and more and more attention is paid to car insurance when buying a car.When there are problems in a certain link of auto insurance,search engines usually cannot give a good answer.Therefore,it is of research and learning value to research and design an intelligent question and answer system in the field of auto insurance to help people solve the problems of auto insurance.In this thesis,deep learning model is used to conduct in-depth research on question matching module based on Chinese vehicle insurance data,aiming to build a question answering system with high accuracy in the field of vehicle insurance.The main research contents of this paper are as follows: First of all,in view of the fact that there is no open source Chinese corpus in the field of vehicle insurance at present,this thesis extracts answers to Chinese questions related to vehicle insurance from the open source Insurance QA corpus,establishes a question-and-answer library of vehicle insurance,and expands the question sentences in the question-and-answer library with positive and negative examples.Then,a data set for training,testing and verification is constructed by manual annotation.Inverted index technology is used to build an index database for the question and answer database,so that when users input questions into the system,the system can use the index database and combine with BM25 algorithm to build a candidate question set for the question.Secondly,the text matching model of ESIM is improved,and the Transformer-ESIM model for question matching is proposed.The improvements to the model are as follows: In the word embedding stage,Transformer's selfattention mechanism is added to solve the problem that the common word vector can only extract the semantic and grammatical information of the question,but cannot obtain the input position information.At the same time,Bi-LSTM unit in ESIM input coding layer and local inference layer is changed into Transformer unit to solve the problem that ESIM cannot obtain the connection between word contexts and retain the high-level semantic information of sentences.Finally,the improved question matching model Transformer-ESIM is applied to the construction process of question answering system to realize an intelligent question answering system of vehicle insurance with high accuracy.This thesis uses Python as the development language and builds a Transformer-ESIM model for question matching with the Tensor Flow deep learning framework.Compared with the benchmark model ESIM model and common text matching model,the accuracy of question matching calculated by Transformer-ESIM model reaches 92.5%,which is higher than the benchmark model and common model,and is greater than 3.3%.This proves that the model can better solve the question matching problem.