Chinese medical named entity recognition of long text based on deep learning

Abstract

Abstract In the era of information technology, the electronic medical system generates vast and diverse data on a daily basis. Maximizing the utilization of this data will have a profound impact on clinical decision-making and public health. Currently, named entity recognition technology has reached maturity in the field of English. However, due to the scarcity of corpus and the complexity of semantic boundary recognition, Chinese named entity recognition remains an area that warrants further research efforts. Based on the classical model of ALBERT-BiLSTM-Self-Attention-CRF, we propose a named entity recognition model named DWI-Pos which integrates the Pos (position information) of entity words and the features of POS (part of speech), and use a DWI (Dynamic Windows Interception mechanism). Subsequently, two comparative experiments were conducted on the BERT-CRF and LSTM-CRF models, and their effectiveness was validated by the results obtained. Furthermore, the entity word position information designed in this study played a significant role in improving the performance of the proposed model. The dataset employed in this study was derived from the sub-tasks of CCKS2019, which includes named entity recognition of Chinese electronic medical records. This task encompasses two sub-tasks, namely medical named entity recognition and medical entity and attribute extraction. The focus of this research was on medical named entity recognition, where the achieved F1 value was 0.95, surpassing the F1 value of the ELMo-ET-CRF model by 0.09.