基于自然语言处理的临床合理用药知识图谱构建

目的：构建基于自然语言处理的临床合理用药知识图谱。方法：以国家食品药品监督管理总局（CFDA）、美国食品药品监督管理总局（FDA）及某大型三甲医院药品库中药品说明书为数据源，构建了一种基于深度学习算法的临床合理用药知识图谱库。对随机抽取的500份药品说明书进行人工标注，将标注的数据划分为训练集、测试集、验证集。基于深度学习模型BRET进行训练，通过训练集训练模型和验证集验证训练过程中的性能及训练后通过测试集测试模型性能，用优化后的机器学习模型预测未标注的药品说明书。结果：最终抽取出30余万条“实体-关系-实体”的三元组关系，将机器学习模型产生的三元组与领域专家标注产生的三元组一起导入Neo4j图形数据库中存储，以知识图谱的形式展现给临床药师。结论：通过基于深度学习算法的临床合理用药知识库构建，在标引少量药品说明书的前提下，挖掘出药品说明书中所有的医疗关系和实体。自动构建基于药品说明书的合理用药知识图谱，可提高合理用药的自动化程度和准确度，降低不合理用药。

Construction of knowledge graphs for rational use of drugs in clinical practice based on natural language processing

Objective To construct the knowledge graphs for rational use of drugs in clinical practice based on natural language processing. Methods A repository of knowledge graphs for rational use of drugs in clinical practice was developed based on in-depth learning algorithm with the introductions of drugs approved by CFDA, FDA and those stored in a certain tertiary class hospital as its data source. The introductions of 500 randomly selected drugs were manually indexed. The indexed data were divided into training set, testing set, and verifying set which were trained based on the in-depth learning model BRET.The BRET model was trained with the training set, the performance of BRET model during the training was verified with the verifying set and tested with the training set after training. The introductions of drugs not indexed were predicted according to the optimized machine learning model. Results The "entity-relationship- entity" triples were extracted from 300 000 introductions of drugs. The triples generated in machine learning model and those generated during the indexing by machine learning experts were input into the Neo4j graphs for storage, and presented to the clinical pharmacists in the form of knowledge graphs. Conclusion Construction of knowledge graphs for rational use of drugs in clinical practice based on in-depth learning algorithm can mine all medical relationships and entities in introductions of drugs by indexing a small number of introductions of drugs. Conclusion Construction of knowledge graphs for rational use of drugs in clinical practice based on introductions of drugs can improve the automation and accuracy of knowledge graphs for rational use of drugs in clinical practice and reduce the irrational use of drugs.