一种在线医疗社区问答文本实体识别方法——基于卷积神经网络和双向长短期记忆神经网络  

An Online Medical Community Q&A Text Entity Recognition Method: Based on CNN and BiLSTM

作　　者：廖开际[1] 邹珂欣 席运江[1] Liao Kaiji;Zou Kexin;Xi Yunjiang(School of Business Administration,South China University of Technology,Guangzhou 510641,China)

机构地区：[1]华南理工大学工商管理学院,广东广州510641

出　　处：《科技管理研究》2021年第8期173-179,共7页Science and Technology Management Research

基　　金：国家自然科学基金项目“基于超网络的企业微博知识挖掘及整合方法研究”(71371077)。

摘　　要：针对在线医疗社区问答文本复杂程度高、结构化程度低的特点,结合卷积神经网络(CNN)和双向长短期记忆神经网络(BiLSTM)两种深度学习模型以及条件随机场(CRF)模型,提出一套适用于在线医疗问答文本的实体识别方法并进行验证。将问答文本进行清洗和BIO标注后,分别用CNN和BiLSTM进行字级别的特征抽取,将通过两种模型抽取到的特征进行融合后放入CRF中训练出实体预测模型,再将问答文本放入训练好的模型中,得到最终的实体识别结果。以关于乳腺癌疾病问答文本为例,研究结果表明,运用该方法得到的识别结果优于其他模型,且识别准确率达到92.3%、召回率达到89.3%、F值达到90.8%。In response to the characteristics of high complexity and low structure in the online medical community Q&A text, this paper proposes and verifies an entity recognition method combined with two deep learning models of convolutional neural network (CNN),bi-directional long short-term memory (BiLSTM) and conditional random field (CRF), to promote the development of medical entity identification research for the online medical community. After the Q&A texts are cleaned and BIO labeled, feature extraction is respectively performed in word-level by CNN and BiLSTM, then the features are fused and the results are put into the CRF to train the entity prediction model, finally the question and answer text are put into the trained model to get the final entity recognition result. Taking the question-and-answer text on breast cancer as an example, the results show that the recognition results obtained by this method are superior to those of other models, and the recognition accuracy rate reaches 92.3%, the recall rate reaches 89.3%, and the F value reaches 90.8%.

关 键 词：实体识别 深度学习 卷积神经网络 双向长短期记忆神经网络 条件随机场 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术] F224[自动化与计算机技术—计算机科学与技术]