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

In recent years,the basic science and cutting-edge technology is developing dramatically,Internet technology can be seen in all aspects of human life.The popularity of the Internet allows people to obtain all kinds of information needed in daily life,which improve the efficiency of people’s life and work.At the same time,the geometric growth of Internet information has also brought huge troubles for people to quickly obtain information.The search engine that emerged at the historic moment provides users with convenient knowledge acquisition channels and meets the needs of users’ information retrieval to a certain extent.For related questions in a specific field,traditional search engines cannot technically return accurate and concise answers.Users still need to screen the answers or ask professional customer service personnel,which undoubtedly wastes the time of users and staff.The development of big data technology and in-depth research on artificial intelligence technology is going to the peak,which makes it possible that the emergence of question answering systems.Many companies and research institutions have begun to conduct in-depth research,and have successively launched question and answer system products,such as Baidu’s Xiaodu Robot,i Flytek’s AIUI platform,Microsoft’s Xiaobing and Xiaona,etc.However,the application in the field of financial is almost blank,and fans in the financial field have become difficult to retrieve relevant knowledge.Based on the forward-looking sub-project of a domestic bank,this article has done an in-depth study on the financial question-and-answer system and related technologies.This paper designs a financial question answering system based on the knowledge graph retrieval and similar question matching of the Neo4 j graph database.First,the question is preprocessed with error correction and colloquial elimination,and then the questions are classified.According to the classification results,they will flow to different paths.When a question is classified into another category,it will be sent back to the user with a safe answer : "Sorry,I can’t answer this type of question yet".When the question is classified as a small chat type,the user will be answered through the small chat interface.When a question is classified as a domain-specific question,the common question library will be searched according to keywords,and the recalled candidate questions will be sorted by similar question matching,and the top three candidate answers with matching scores will be returned.Secondly,a knowledge map search will be performed,and entity relationship information will be extracted from the user’s question before the knowledge map search is performed.This article starts with entity relationship extraction,uses entity relationship extraction for knowledge extraction,and proposes an end-to-end joint entity relationship extraction model based on deep neural networks.The experimental results prove the effectiveness and superiority of the model.Through the extracted entity relationship information,fill in the cypher sentence template,get the answer from the knowledge graph,and finally return the best answer.The experimental test results show that the question answering system in this paper is scientific and effective.