Chinese Electronic Medical Record Named Entity Recognition based on FastBERT method

Abstract

Chinese Electronic Medical Record Named Entity Recognition (CNER) is to identify and extract the entities related to medical and clinical practice from electronic medical records and classify them into pre-defined categories. In the past few years, deep learning methods have been applied to CNER and have achieved remarkable results, especially the BERT pre-training model. the BERT model can achieve good results, but the high model's training cost and slow inference speed are unbearable. In order to solve these problems, scholars use various methods to compress the BERT model, such as knowledge distillation and architecture adjustment. In this article, FastBERT is improved and applied to CNER. The sample adaptation mechanism of this model is used to pick up the inference speed. It is learned from experiments that this method can not only improve the reasoning speed of entity recognition, but also maintains good performance.