Chinese electronic patient record entity identification method based on GPT-2 model

• Main Authors: ZHU GUOSHENG, QI XIAOYUN, WU MENGYU, LIU FEIHONG, WU SHANCHAO
• Format: Patent
• Language: Chinese; English
• Published: 10.01.2020
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention relates to a Chinese electronic patient record entity identification method based on a GPT-2 model. The method comprises the steps of extracting the feature vectors of the electronic patient records by utilizing a GPT-2 pre-training model, obtaining an identification probability by taking a CRF model as an outlet, and finally obtaining the named entities of the Chinese electronic patient record. The method comprises the steps of dividing the data of the Chinese electronic patient records into a training set and a test set, and labeling the data of the two parts in a unified mode,wherein the labeled data comprises an original Chinese electronic patient record and an entity label; 2) on the basis of a GPT-2 pre-training model, introducing the CRF model, establishing a GPT2-CRF-based Chinese electronic patient record entity recognition model, and training by using the data of the training set to obtain a trained Chinese electronic patient record entity recognition model; and 3) inputting the data o