基于深度学习的患者安全事件的命名实体识别

Named entity recognition of patient safety events based on deep learning

为解决日益增长的数据体量与实际作业效率和成本的矛盾,用科学的标注体系对患者安全事件语料进行标注,设计了一种基于深度学习的BERT-BiLSTM-CRF模型,结合中文文本语料的语义特征和字符特征对其进行命名实体识别,最终实验F1值为91.49%,相较于BiLSTM-CRF模型和IDCNN-CRF模型,实体的识别性能分别提升了7.33%和8.30%,验证了该模型的有效性。

A deep learning-based Bert-BiLSTM-CRF model was designed by annotating the patient safety events-related language corpus features using the scientific annotation system in order to solve the contradictions of the increasing data volume with the actual operation efficiency and cost.The named entity was recognized using this model in combination with the semantic features and character features of Chinese text,which showed that the F1 value was 91.49%.The recognition performance of this model increased 7.33%and 8.30%respectively than that of Bert-BiLSTM-CRF model and IDCNN-CRF model,thus verified the efficiency of this model.