基于网络表示学习的疾病知识关联挖掘与预测方法研究  

Disease Knowledge Association Mining and Forecasting Based on Network Representation Learning

作　　者：余黄樱子 董庆兴 张斌[4,5] 

机构地区：[1]武汉大学信息管理学院,湖北武汉430072 [2]武汉大学信息资源研究中心,湖北武汉430072 [3]华中师范大学信息管理学院,湖北武汉430079 [4]武汉大学中国传统文化研究中心,湖北武汉430072 [5]武汉大学国家文化发展研究院,湖北武汉430072

出　　处：《情报理论与实践》2019年第12期156-162,共7页Information Studies:Theory & Application

基　　金：国家自然科学基金面上项目“面向群智感知大数据的群体评价模型与方法研究”(项目编号:71871102),国家自然科学基金国际(地区)合作与交流项目“基于慢病知识管理的智慧养老平台研究”(项目编号:71661167007)的成果;中国科协青年人才托举工程项目“信息技术与管理”(项目编号:2018QNRC001)

摘　　要：[目的/意义]近年来,以电子病历大规模应用为标志的医疗信息化发展迅速.电子病历的广泛应用使得医疗信息管理机构产生并存储了大量医疗数据,如何从这些海量的数据中挖掘出有价值的疾病关联知识,以辅助疾病的精准诊断和预测成为当前信息科学研究人员和业界人员面临的重要问题.[方法/过程]文章提出一种疾病知识网络表示学习模型及其链路预测算法(NRL-LP),以进行疾病知识的关联关系挖掘与预测.该模型学习网络节点的内部和外部特征并将节点映射为空间向量以浅层表示节点,然后将链路预测问题转化为有监督的学习,提出NRL-LP算法预测节点对之间是否连接来挖掘和预测疾病知识间存在的关联关系.[结果/结论]以1400万条非结构化医疗临床记录的数据集为实验对象,结果表明,NRL-LP能够揭示新的疾病关联知识,为有效进行临床决策提供帮助.[Purpose/significance]Recently,health information technology marked by extensive implementation of electronic health records has developed rapidly.The widespread adoption of electronic health records allows medical information management agencies to generate and to store an enormous amount of medical data.How to extract the relationships between diseases from these medical data for the accurate diagnosis and prediction of diseases has become an important issue for both information science researchers and industrial professionals.[Method/process]In this paper,representation-learning model of disease knowledge network and its link prediction algorithm(NRL-LP)is proposed.This model combines nodes,internal and external features to encode the nodes in an n-dimension space through network representation learning to obtain the latent representation of the nodes.Then,link prediction is considered as a kind of supervised learning.It will learn to distinguish the connected cases(connected and unconnected edges)in the network to predict the likely connected node pairs(the connected edges indicate that the disease knowledge nodes may be associated)to complete the disease knowledge association mining and forecasting.[Result/conclusion]This study makes an extraction of 14 million unorganized medical clinical records.Experimental results show that the proposed algorithm can discover and predict disease associations precisely to help clinical decision making effectively.

关 键 词：知识挖掘 链路预测 电子病历 预测方法 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]