Research And Implementation Of Knowledge Answering System In Insurance Field

In recent years,with the rapid development and popularity of the Internet,all walks of life have begun to provide online platforms to complete communication and transactions with users.In order to answer the user's doubts in a timely manner and improve the user experience,the service provider usually provides a manual customer service platform to communicate with the customer,thereby avoiding unnecessary economic losses.However,in the face of a large number of online users,manual customer service will inevitably encounter a large number of repetitive problems and consume a lot of unnecessary human resources.Therefore,a knowledge-based question answering system that can automatically communicate with the user is especially important.In the process of constructing the knowledge-based question answering system,the user input question needs to be mapped into the existing knowledge base questions,and the classification of the question can reduce the number of candidate questions and improve the response efficiency of the system.Therefore,this paper studies in depth on the topic of classifying the question.Based on the research results of common classification models,this paper investigates and implements three commonly used neural network classifiers:MLP(Multilayer Perceptron)classification model,CNN(Convolutional Neural Networks)classification model and LSTM(Long Short-Term Memory)classification model.Then,based on the question corpus in the insurance field,a combined classification model is proposed.Using the linear regression method,the three mature classification models of MLP,CNN and LSTM are combined to construct a combined classification model for the system.Finally,based on the corpus test set and the artificially designed variant question test set,some experiments are designed to compare the classification effect of each classification model,so as to improve the accuracy of the question classification model.Based on the research status of the knowledge-based question answering systems and the research result stated above,this paper uses common development framework Spring,SpringMVC,MyBatis to implement a knowledge-based question answering system using insurance domain corpus.The question answering system designed and implemented in this paper,after receiving the input question from the user,will perform three-step operation of question understanding,question classification and question matching on the input question,to match questions in the established knowledge base to find similar question and response the answer to the user.If the question answering system cannot find a suitable answer in the knowledge base,it will notify the staff to perform a manual response,and then add the new question-answer pair into the database,so that the system can continuously improve and self-grow.In the question understanding module,the methods of understanding the implied information of the question are summarized from multiple dimensions.The research realizes the functions of various question understanding,including:word segmentation of Chinese questions,training of word vector models,identification of the direction of the question,the judgment of the question type,the analysis of the sentence structure,the extraction of the main clause,the supplement of the context,and so on.In the question matching module,considering the unique environment of the system,the question matching model based on the question understanding result is designed and implemented to complete the work.