计算机技术与发展2019，Vol.29Issue(2)：106-108,142,4.DOI:10.3969/j.issn.1673-629X.2019.02.022
基于LSTM-CRF命名实体识别技术的研究与应用
Research and Application of Named Entity Recognition Based on LSTM-CRF
摘要
Abstract
With the development of deep neural network, deep learning not only occupies the dominant position in pattern recognition and other fields, but also has been applied to various aspects of natural language processing, such as Chinese named entity recognition.When recognizing named entities in electronic medical records, we construct a long and short time neural network model with embedded random field.The context vector of the hidden layer of long and short time neural networks is used as the feature of output layer annotation, and the embedded conditional random field model to represent the constraint relationship between the annotations.The model identifies five types of entities, including body parts, disease name, examination, symptom and treatment in the electronic medical record, with accuracy of 96.29％, precision rate of 91.61％, recall rate of 96.22％, and F value of 93.85.For the entity category of symptom, the precision rate reaches 96.08％, recall rate of 98.98％, F value of 97.51.The experiment shows that the proposed model is effective in identifying Chinese named entities, which is helpful for the automatic extraction of the relationship between entities in Chinese electronic medical records and the construction of medical knowledge maps.关键词
长短时记忆神经网络/条件随机场/命名实体/电子病历Key words
long and short time memory neural network/conditional random field/named entity/electronic medical record分类
信息技术与安全科学引用本文复制引用
张聪品,方滔,刘昱良..基于LSTM-CRF命名实体识别技术的研究与应用[J].计算机技术与发展,2019,29(2):106-108,142,4.基金项目
河南省科技攻关项目(152102310313) （152102310313）
河南师范大学专业学位研究生课程案例库建设项目(5101119500706) （5101119500706）
