Research On Multi-source Heterogeneous Knowledge Fusion For Smart Health

Smart health is a brand new medical care model.Effective acquisition,organization,query and analysis of the massive,heterogeneous and multi-source health knowledge resources are the key to realize the “smart” health care.Combined with users' health knowledge needs and the characteristics of smart health knowledge,this paper focused on the field of smart health,aimed to extract,transform,evaluate and integrate knowledge units from multi-source heterogeneous smart health knowledge resources(such as health care knowledge,medical study,clinical diagnosis and treatment experience distributes in the Internet,scientific literature database,specialist diagnosis and treatment data sets and other places),through social network analysis method,machine learning method and semantic Web technology,etc.Finally,a large-scale smart health knowledge graph was formed to solve specific problems,and provide intelligent and efficient health knowledge services for users.In this process,it mainly includes the following specific work.Firstly,the definition and connotation of knowledge fusion were discussed,and related concepts were discriminated.Then Knowledge fusion can be defined as:Geared to the needs of users or specific field problems,knowledge from different sources and their dependent carriers are extracted and transformed through certain methods and technical means.Then knowledge units and relationships hidden were acquired,assessed,analyzed and fused at the conceptual level and semantic level,to form a knowledge base or knowledge graph that can solve specific domain problems and provide users with more intelligent knowledge services.Secondly,a multi-source and heterogeneous knowledge fusion model and service architecture for smart health was constructed.According to the definition and sources,smart health knowledge can be divided into disease standard documents,biomedical literature,medical network resources,structured ontology and specialized diagnosis and treatment data sets.Then,based on the general public's demands for smart health knowledge services under the big data environment,tasks and challenges faced by the knowledge fusion of smart health were analyzed,as well as the process,goal and realization path.On this basis,a multi-source heterogeneous knowledge fusion model and service architecture for smart health was proposed.Thirdly,the explicit and implicit health knowledge needs of users in online health communities were explored.We first analyzed the types,hierarchy and evolution patterns of user's health knowledge demands,and then the typical patientpatient health community Patients Like Me and Manyoubang were taken as examples to explore the health knowledge needs of patients with hypertension at home and abroad.In order to excavate the explicit health knowledge needs of users,topic model HDP was adopted to mine the potential topics in the user's post content,then the KL distance,content confusion,and model complexity among topics obtained by the HDP model and the LDA model were compared and discussed.Research found that etiology,treatment,drugs,diagnosis,symptoms are of common interests to users at home and abroad,as well as the explicit health knowledge needs expressed by users.On the other hand,in order to excavate user's implicit health knowledge needs,an exponential random graph model was used to analyze the influencing factors and formation mechanism of the patients communication network.Combining the basic information and behavior patterns of patients,users' implicit knowledge needs in the objective stage and consciousness level were explored.This method provides a new way to study the users' implicit health knowledge needs.Fourthly,the extraction task,process and method of different types of smart health knowledge were explored.Such as the structured medical ontology or knowledge base,semi-structured web encyclopedia data and unstructured free documents were summarized and illustrated with examples.Then,taking the biomedical literature related to hypertension in Pub Med database as an example,the hierarchical structure and term association of Me SH terms co-occurrence network were analyzed by traditional tomographic content analysis method.At the same time,deep learning method Bi LSTM-CRF and Att-Bi LSTM model were utilized to extract entity and relationship from large-scale biomedical literature.On this basis,topics in the entity-relationship maps was excavated and semantically annotated to prepare for the next step of knowledge representation and fusion.Fifthly,the fusion process and implementation path of multi-source heterogeneous smart health knowledge were explored.On the one hand,knowledge(triples)extracted from multi-source heterogeneous health information resources were fused into the top-level ontology of the smart health domain through entity fusion,attribute fusion and concept fusion at the conceptual level.On the other hand,all kinds of medical care knowledge extracted from multi-source heterogeneous health information resources were represented as ontology or knowledge base through the process of ontology blocking,ontology alignment and entity matching,to realize the combination of ontology,knowledge base and knowledge graph at the semantic layer.Then,taking hypertension as an example,this paper fused the top-level ontology(constructed based on the Disease Ontology and the guidelines for the prevention and treatment of hypertension)and the field experience ontology(constructed based on the topics in the entity-relationship map extracted from the large-scale biomedical literature)to explore the process and realization path of multi-source heterogeneous knowledge fusion.Research found that the fused hypertension domain ontology has a more comprehensive conceptual system,richer ontology content and more diverse domain knowledge categories.Finally,the smart health knowledge graph was constructed to provide knowledge services for users.Health knowledge needs of user should matched with the health knowledge resources.On the basis of the top-level ontology in the field of hypertension,knowledge resources extracted from multi-source heterogeneous health information were fused to form a knowledge graph to solve specific problems.Embed it into the smart health knowledge service platform,it can provide users with knowledge visualization,knowledge retrieval,knowledge recommendation,decision support and other smart health knowledge services.This paper consists of 8 chapters,including 65 figures,23 tables and 270 references.