Research On Construction Of Financial Events Knowledge Graph And Its Application

In recent years,Internet technology has entered thousands of households,and all walks of life are increasingly dependent on the Internet.Due to the huge scale of Internet users and the convenience of obtaining information from the Internet,many enterprises and financial institutions regard the Internet as one of the preferred media for information disclosure.The financial events hidden behind these information will have a significant impact on capital market investment and investors' decisions.Financial information on the Internet is complicated,and it is very difficult for financial practitioners to obtain the key elements and relationship of events.The Financial Knowledge Graph provides a very convenient way for financial practitioners to access information about financial markets.However,the current research on financial knowledge graph is centered on financial entities such as enterprises and corporate legal persons.Such financial knowledge graph tends to weaken the influence of event relationship on financial entities.This paper will build a knowledge graph with financial events as the core to solve this problem.In order to construct the knowledge graph of financial events,event extraction is needed for financial announcements.At present,the existing researches on event extraction are mostly used to extract the event elements in a certain sentence,but the event elements in financial announcements are usually dispersed to a large extent,so it is necessary to make full use of the information of the whole document.This paper first encodes the financial announcement and obtains the word encoding vector of the text.In order to make use of global context information in entity extraction,sentence coding and word coding are combined to facilitate the information flow between all entity coding and sentences.Event trigger classification is carried out after the entity encoding at the chapter level is obtained,and then the directed acyclic graph of financial events is constructed by using the classifier to complete the event extraction of financial announcements.Social affective orientation towards enterprises plays an auxiliary role in risk prediction,but the current financial knowledge map does not pay attention to social affective orientation.This paper adds corporate emotional results to enrich the knowledge graph of financial events.At present,most of the sentiment analysis algorithms are oriented to the social field or the general field,but the performance in the financial field is not good.In this paper,Finbert is used to code financial news,and then aspect based sentiment analysis method is used to conduct sentiment analysis on financial news.At the same time,in order to enhance the accuracy of financial risk prediction,this paper collects enterprise financial index data as auxiliary data.Using the research results of financial management,this paper calculates the five abilities of enterprises and obtains the score value of each ability.Finally,this paper constructs the event knowledge graph with financial events as the core to enhance the influence of the connection between financial entities on financial risk prediction.The emotional results and financial capabilities of financial news are added to the knowledge graph through entity fusion.Then,the analysis results of the three data sources were fused to build a time series prediction model and three features were obtained.These three features were classified by Softmax with attention mechanism to predict the financial risk of the enterprise,and the final prediction accuracy was 75.52%.