【作者】 刘雨；

【导师】 臧天仪；

【作者基本信息】 哈尔滨工业大学 ， 计算机科学与技术， 2022， 硕士

【摘要】 随着信息技术的发展,医学健康数据呈指数级爆炸式增长,数据变得越发庞大冗杂,医疗健康术语或概念实体之间的关联分析利用面临巨大挑战。针对此方面问题,本文在建立医疗健康知识图谱的基础上,利用特征学习的方法实现基于特征空间的医疗健康知识图谱检索和医疗健康文献询证检索。本文的主要工作有:(1)研究大规模医疗健康领域知识图谱的构建方法。从数据采集处理、知识抽取以及实体消歧等方面开展深入研究并构建医疗健康知识图谱,构建基于MIMIC-III数据集的医疗健康知识图谱,为医疗健康领域相关概念检索提供数据基础。(2)研究基于特征空间的大规模医疗健康领域知识图谱检索算法。通过建立医疗健康知识图谱特征邻域,挖掘发现潜在图谱实体关联关系,采用特征空间上的一个领域表示候选节点集合,通过使用医疗健康知识图谱作数据集,分析不同知识图谱检索方法,验证算法的有效性及科学性。(3)研究基于特征空间的文献循证检索方法。利用医疗健康实现与文献询证关联特征实现查询筛选,通过强化学习模型学习适应用户需求特征,发现询证文献,从而更好的捕捉用户的兴趣偏好,通过模拟实验验证方法的合理性。本文在医疗健康知识图谱的检索上提供了一种新的方法,能够支持大规模医疗健康知识图谱查询以及文献询证检索,具有重要的科研和临床应用价值。更多还原

【Abstract】 With the development of information technology,medical health data is growing exponentially and explodes.The data is becoming more and more complicated,and the association analysis and utilization between terms or conceptual entities of medical health is facing great challenges.Aiming at this problem,based on the establishment of medical health knowledge graph,this paper uses the method of feature learning to complete the retrieval of medical health knowledge graph and medical health literature based on feature space.The main work of this paper is as follows:(1)Study the construction method of large-scale knowledge map in medical health field.In-depth research was carried out from data acquisition and processing,knowledge extraction and entity disambiguation,and medical health knowledge map was constructed.The medical health knowledge graph based on MIMIC-III dataset was constructed to provide data basis for relevant concept retrieval in medical health field.(2)Research on the feature space based large-scale medical health domain knowledge graph retrieval algorithm.By establishing the characteristic neighborhood of the medical and health knowledge map,the potential entity association relationship of the graph was discovered,and a field in the feature space was used to represent the candidate node set.By using the medical health knowledge graph as the data set,different knowledge graph retrieval methods were analyzed to verify the validity and scientific nature of the algorithm.(3)Study the literature evidence-based retrieval method based on feature space.The association features of medical health implementation and literature verification are used to realize query and screening.The reinforcement learning model is used to learn and adapt to users’ interested characteristics,and the literature verification is found,so as to better capture the interests and preferences of users.The rationality of the method is verified by simulation experiment.This paper provides a new method for the retrieval of the medical health knowledge graph,and combines the reinforcement learning technology to retrieve the target literature more quickly.更多还原