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

In recent years.with the fast-improving of the social economic.it has led to a growing enthusiasm for investment.Investment products such as funds,financial management,stocks and Peer to Peer Lending(P2P)are becoming more and more popular with consumers.Each investment product contains a variety of different services,which makes the business and product too complex.It requires investors to have a detailed understanding of the product before buying it.However,there are few ways to understand product information for the investor.Generally.they may consult the relevant staff first.and then get more information through the search engine on the Internet.But it is worth noting that communication with relevant staff may is usually a long process,such as service reservation,which can cost investors a lot of time.At this time,the search engine has attracted more and more people's attention because of its high efficiency and convenience.However,some shortages of current search engine technology cannot be ignored.Specifically.the Search engine is a search mode in all fields.It is the wide range of characteristics determines that it cannot satisfy all the demands of customers,and can only answer some basic questions.and the accuracy of the answers is not high.and the quality is uneven.Thus.the demand for an online question answering system that can accurately understand the user's questions and support them a correct and objective answer to the user has become more and more intense.Traditional question answering systems are generally designed and implemented based on machine learning algorithms.This model has three major shortcomings.Initially,word meaning and semantics need to be entered manually.Secondly,insufficient growth of the model.Thirdly,there are not deep enough in understanding user semantics.We will divide this research into three parts.Firstly.finding the Word Vector Model of the Question Answering System for the Most Consistent Alloys.In order to express word meaning more accurately,CBOW model in Word2vec is selected to train word vector.and compared with TF-IDF.One-hot and other word vector representation methods to verify the superiority of CBOW model.Furthermore,based on CBOW word vector model,and propose the F-B question-and-answer model.In order to deepen the understanding of core questions and reduce redundant information,attention mechanism is used to refer to the model to make the final semantic understanding vector more characteristic.In order to make the model have a good memory function for the context of the text,Bi-LSTM is introduced in this research.In order to integrate the amount of information and retain the key information.Max-Pooling is introduced to integrate these technologies and propose an F-B model.Then the comparative experiments with support vector machine(SVM),convolutional neural network(CNN)and other models are carried out to'verify the superiority of the F-B model in the financial field.Finally,based on F-B question-and-answer model,a complete financial question-and-answer system is designed and implemented.The system consists of three modules:the data processing module,the question understanding module and the iteration update module.