Research And Implementation Of Event Extraction System For Equity Change

As a crucial portion of information extraction,event extraction can extract specific event triggers and related arguments from unstructured text.In the financial field,obtaining the equity change events from the financial announcement can help people understand the changes of the company’s equity.The financial declaration deals with the operating conditions of listed companies.When the equity increase,decrease,freeze,repurchase and pledge events occur,it will affect investors’ decisions.Extracting events from financial announcements faces some challenges.The arguments of events in financial announcements are usually scattered throughout the whole document.It is essential to comprehend the knowledge in the whole document,and extract all arguments across sentences by combining the information of multiple sentences.This challenge makes the existing event extraction methods for general fields have low effectiveness.This paper invents an extraction and self-service inquiry system of equity change events to solve the above problems.The system has efficiently accelerated the speed of financial practitioners in obtaining information about equity change events from financial announcements,and ordinary people can more conveniently know the information about equity change of interested companies.The main work is as follows:(1)An equity change event extraction system has been implemented.The system will provide users with automatic extraction and self-service inquiry services of equity change events,regularly obtain the latest announcement of equity change events from the Internet and extract the equity change events contained therein to update the database.Users can come into the page across the browser and enter the company name,and obtain the information of equity change events related to the company,or upload the existing equity change announcement file to the system for online extraction and instantly obtain the information of equity change events contained in the announcement;(2)For purpose of implement the capabilities of the system,a document-level event extraction model is proposed.By constructing a directed acyclic graph,the model can extract multiple events and arguments spreaded in the document.The global memory unit is used to realize the model’s perception of the current path state and historical context.The gated context attention layer enables the entity vector to obtain the context information across sentences in the document,thus effectively realizing the event extraction for the announcement of equity change events.In this paper,the public dataset constructed by large-scale financial announcements is used to verify the model,and the devised model has achieved excellent performance,which certifies its efficiency.