Design And Implementation Of Question Answering System In Finance Based On Event Evolution Graph And Knowledge Graph

With the development of information science,various industries have gained new vitality driven by data science.The financial industry,as a highly correlated industry,is also facing rapid changes in digitalization and intelligence.Knowledge graph technology is the product of the combination of artificial intelligence and traditional databases technology.It uses information extraction technology to extract human-focused knowledge from unstructured text and store it in the form of knowledge graph.In the financial field,data resources not only contain "static" knowledge such as corporate information and personnel information,which can be covered by traditional knowledge graphs,but also include "dynamic" knowledge centered on events.The event evolution graph technology is concerned with events and the logical relationship between events.It is an extension of the knowledge graph technology in "dynamic" knowledge.Therefore,the event evolution graph technology can be used to manage the "dynamic" knowledge in the financial field.The knowledge base-based question answering system has the advantages that traditional information retrieval systems do not have.It can handle questions in line with human oral expression habits.Using graph technology as a support,it can automatically construct knowledge bases and ensure knowledge bases updated rapidly.Therefore,we build an event evolution graph and a knowledge graph for the financial industry.Considering the availability of knowledge at the same time,we have constructed a knowledge-based question answering system based on knowledge graph and event evolution graph.The main work of this thesis includes:(1)Aiming at the management of event knowledge in the financial field,this paper proposes a method for constructing event evolution graphs.This method uses the deep neural network model to automatically extract text features,avoiding manual selection and construction of features.The method uses sequence labeling algorithm to extract causal event pairs,effectively exploring hidden event causality.It uses deep embedding clustering algorithm for event clustering to improve accuracy.And finally,it uses results of event clustering to construct event evolution graphs.We conducted experiments on real financial news corpus.The experiments show that our method has achieved good results in accuracy,recall and F-measure.(2)This thesis designs and implements a knowledge-based question answering system based on the event evolution graph and knowledge graph.From the perspective of software engineering,this thesis clears the requirements of the system and carries out the overall and detailed design of the system.The system is divided into three modules: construction of event evolution graph,construction of knowledge graph and question and answer based on knowledge base.In this thesis,according to the characteristics of knowledge in the financial field,based on information extraction and deep learning related technologies,we construct an event evolution graph and a knowledge graph,and a question and answer system based on the event evolution graph and knowledge graph is implemented.Finally,we tested each module of the system.The test results show that the system has met the system design requirements and goals.In general,this thesis proposes a method for constructing event evolution graph based on deep sequence annotation and deep clustering algorithm in the financial field.Based on this algorithm,a question answering system based on knowledge graph and event evolution graph is designed and implemented.