基于深度学习的名老中医用方经验传承知识图谱研究

Study on Knowledge Map of Prescribing Experience Inheritance of Veteran TCM Physicians Based on Deep Learning

目的构建名老中医经验传承知识图谱,为名医学术经验传承和创新提供依据。方法整理2019年9月16日-2021年3月22日张永康教授于山西省人民医院中医科门诊部分病历,通过人工标注生成数据集,按7∶3将其随机分成训练集(n=804)和测试集(n=345),采用TPLinker模型进行实体关系联合抽取,使用精确率、召回率、F1值作为模型的评价指标,通过Py2neo和Neo4j构建中医医案知识图谱,并使用Cypher查询语句实现知识图谱的可视化过程。结果TPLinker模型在中医医案数据集上的精确率为0.886、召回率为0.977、F1值为0.929。通过Py2neo库构建了包含1025个实体、4428条关系的中医医案知识图谱,并将其存储在Neo4j图数据库中。结论名老中医用方经验传承知识图谱能充分挖掘张永康教授个性化辨证思维模式,为传承和发扬名老中医学术思想及临床经验提供新途径。

Objective To construct a knowledge map for the experience inheritance of veteran TCM physicians;To provide theoretical basis for the inheritance and innovation of academic experience of famous doctors.Methods Professor Zhang Yongkang’s partial medical record data from the outpatient department of TCM of Shanxi Provincial People’s Hospital from September 16,2019 to March 22,2021 were sorted out,and a dataset was constructed through manual annotation,and then were randomly divided into a training set(n=804)and a test set(n=345)at 7:3.TPLinker model was used for entity relationship joint extraction.Precision,Recall,and F1 value were set as the evaluation indicators of the model.Py2neo and Neo4j were used to construct a knowledge map of TCM medical records,and Cypher query sentences were used to realize the visualization process of knowledge map.Results The Precision of TPLinker model in the TCM medical record dataset was 0.886,the Recall was 0.977,and the F1 value was 0.929.The TCM medical record knowledge map contained 1025 entities,and 4428 relationships was constructed through the Py2neo database and stored in the Neo4j diagram database.Conclusion Knowledge map for the experience inheritance of veteran TCM physicians can fully tap into Professor Zhang’s personalized thinking mode of syndrome differentiation,providing a new way to inherit and promote the academic ideas and clinical experience of veteran TCM physicians.