Research On Event Extraction For Electronic Health Records

Event extraction can extract valuable structured information from unstructured text for computer to process.Useful information for diagnostic reasoning can be extracted from the electronic health records(EHR)by event extraction.Event representation model and seed introduced joint constraint clustering based EHR event extraction model was designed for EHR.The main tasks are:(1)A trigger and argument role based hierarchical EHR event representation model TARHERM was designed.The framework of TARH-ERM,types of trigger and argument role,and types of event relation was designed suitable for EHR.(2)Taking events represented by TARH-ERM as the target of event extraction,a seeded introduced joint constraint clustering EHR event extraction model was designed.For trigger and argument role recognition,word vector similarity and semantic constraints function are considered to calculate the similarity between triggers and triggers and the similarity between argument roles and argument roles,and seed words are introduced into the clustering process to identify the types of triggers and argument roles.Associations between words representing internal relationships of events and relationships between events are established by using sematic information.(3)With SJCC-EEM as the core,the event extraction prototype system SJCC-EESys was designed and implemented.The main achievement is to propose EHR trigger and argument role based hierarchical event representation model and event extraction model of seed introduced joint constraint clustering.This work has a reference value for the study of medical text extraction.