Information Extraction Study On Patient Safety Incidents In Patient Complaints

[Purpose] This study for Chinese patients with lack of reliable security incident field analysis tools and deep learning method is not applied widely,build scientific information extraction model(including entity recognition and the relationship between entity extraction),for the development of information extraction system lays the foundation in the field of patient safety,this study aims to: formation of patient safety events entity tagging system;Generate the corresponding annotated corpus;The information extraction of patient safety event based on deep learning method is realized.[Methods] The information extraction model based on Bi GRU model was constructed based on 1447 complaint texts related to patient safety incidents collected by a third-grade a hospital follow-up system and We Chat application.Referring to the classification system of patient safety incidents at home and abroad,the labeling system of Chinese patient safety incidents is constructed.BRAT is used to mark text corpus.BERT’s Chinese pre-training model is used to form character vector training for Embedding layer,and the representation of text character vector is obtained;Bilstm-crf model is used to identify entity and event types in text and complete sequence prediction.The attention-based Bi GRU model is used to extract the relationship between entities.[Results] This research reference for the classification system of patient safety events at home and abroad,constructs the related tagging system(eight kinds of entity tag,seven kinds of relations between entities)through the round of annotation and formal notation,1447 incidents involving patient safety in the real world of manual annotation text corpus,forming the tagging corpus of patient safety events;The corpus bert-bilstm-crf was constructed for entity recognition.The average F1 value was 91.49%.Compared with the bilstm-crf model and IDCNN,the recognition performance of entity labels was improved by 7.33% and 8.30% respectively.Attention-based Bi GRU was used for relation extraction,and the average F1 value reached 84.13%.[Conclusions] Based on the deep learning method,this study extracted semantic information such as entities and the relationship between entities in text corpus,realized the information extraction of Chinese patient safety incidents,and laid a certain foundation for the development of information extraction system in the field of patient safety.