《多领域本体的支持如何重用这些本》由会员分享，可在线阅读，更多相关《多领域本体的支持如何重用这些本（25页珍藏版）》请在金锄头文库上搜索。
1、Automatically Ontology Constructed From Online Ontology20081225n数据库的语义检索需要多领域本体的支持。如何 重用这些本体？nWEB上越来越多的InterLinking RDF data, 粗略 估计 ，RDF三元组的数量为：l2 billionlInterlinked by around 3 million RDF links。n当前基于keyword的 语义搜索引擎（swoogle, SWSE, Falcons等）仅返回一个本体列表，并没 有进一步的提供本体重用支持。nWeb上越来越多的本体为用户提供API，如 YAGO，DBP
2、edia,Gename,DBLP等Background and Motivationn本体开发的方式lFrom scratch很困难，需要领域专家的参与lOntology reuse一种好的选择n本体的构建步骤l确定应用领域术语l明确术语间的联系、属性及约束n本体的“模块化”（Modularization）技术允许用 户协同开发，导入本体模块（module）,但并没 有提供抽取（extraction）或合并(merge)用户感 兴趣的本体部分的支持.n如何自动构建面向特定应用的领域本体？Background and MotivationArchitectureOnto ExtractorOnt
3、o RankerExtractorMap column names that are not foreign key can be translated into Datatype Property. Translation from RDB to RDF(S)/OWLnDrawbackslWithout considering more constraint information and records information, example columns value constraint.l Translating all column names that are not fore
4、ign key into datatype property may be not appropriate for some specific application.Translation from RDB to RDF(S)/OWLnSwoogle OntoRanklDifferentially treat the following 4 relations. Each of relations were assigned differently weight.uImports(A,B)uUser-terms(A,B)uExtends(A,B)uAsserts(A,B)Ontologies
5、 rankingnOntokhojlConsidering concepts of referencing, the ontology referencing hyperlinks Ref rdf:type, rdfs:subClass, rdfs:domain, rdfs:range, rdf:seeAlso,rdf:about, owl:sameAs, rdfs:isDefinedBy. Assigning different weight to different links Ontologies ranking(2)nContent based rankingnuses Google
6、search result(WordNet was used to expand user search terms), and take the first N pages as corpus. High tf-idf score terms can be considered as concept labels. nEach ontology is then ranked according to how many of these terms match class labels within them. nClass Match Score(CMS) isOntologies rank
7、ing(3)Ontologies ranking(4)nAKTiveRanklClass Match Measure (CMM)lDensity Measure (DEM)lSemantic Similarity Measure (SSM)lBetweenness Measure (BEM) Definition: Let M =M1,Mi,.M4= CMM, DEM, SSM, BEM,wi is a weight factor,and O is the set of ontologies to rank.Ontologies ranking（5）nClass Match Measure (
8、CMM)lEvaluate the coverage of an ontology for the given search terms.Ontologies ranking（6）nDensity Measure (DEM)lEvaluate degree of detail of searching class. Let S = S1,S2,S3,S4 =relationsc, superclassesc, subclassesc,siblingscOntologies ranking（7）nSemantic Similarity Measure lThe motivation is tha
9、t ontologies which position concepts further away from each other are less likely to represent the knowledge in a coherent and compact manner. Let ci ,cj classeso, and ci cj is a path p P of paths between classes ci and cjOntologies ranking(8)Ontologies ranking(9)nBetweenness MeasurelNodes that occu
10、r on many shortest paths between other nodes have higher betweenness value than others.lThe assumption is that if a class has a high betweenness value in an ontology then this class is central to that ontology.Ontologies ranking(10)Let ci,cj classeso, ci and cj are any two classes in ontology o,Co i
11、s the set of class in ontology o, bem(c) is the BEtweenness Measure for class c.Ontologies ranking(11)nClassificationlTerminologicaluString-based uLinguistic Language-based Linguistic resourcelStructuraluInternal Constraint-based Alignment reuseuExternal Graph-based Taxonomy-baselSemantic umodel-bas
12、edOntology mappingnString-basedl主要思想：将所要匹配的实体的名称划分为一个字母 系列。包括prefix, suffix, edit distance, N-gram等匹配 方法nLanguage-basedl主要思想：充分利用所要匹配的实体名的形态特性和 NLP处理技术。包括分词,词干还原,去除停用词。nConstraint-basedl主要思想：利用实体的定义信息，包括类型，属性的 集势、KEY约束Ontology mapping（2）nLinguistic resourcel主要利用词典、主题词表等进行匹配nAlignment reusel主要是利用先前的匹
13、配结果来指导当前的匹配nGraph-basedl把所要匹配的输入看成是标签图（包含所要匹配的术 语及它们的关系），两个结点的相似性比较是基于它 们在图中的位置，即不同源中的两个结点相似则它们 的邻居在一定程度上也是相似的。nTaxonomy-basedl实质上也是一种图匹配技术，仅考虑词之间的特定关 系（is-a)Ontology mapping（3）nModel-basedlPropositional SATisfiabilityu将匹配问题翻译为命题（formula）表达式，然后检查命题的 有效性。但仅限于一元谓词对于命题的可满足性判断而言， SAT Descider是正确的和完备的lMo
14、dal SATu对SAT的扩展，使其能处理二元谓词。主要思想是用模态逻辑 或ALC描述逻辑增强命题逻辑lDL-basedu以一种纯语义的方式检查类（class）之间的包含（ subsumption）关系。首先合并所要映射的本体，在相同的解 释集下，对所要匹配的每一对概念和关系（role）检查他们之 间的包含关系。如果发现不满足，则重新考虑他们之间的映射 。Ontology mapping（4）nModularlBasic Description LogiclMethoduDecomposing a large and comprehensive ontology into a set of smaller and self-contained modules.uIntroducing new formalisms for developing modular ontologiesnExtraction graph traversalOntology Extraction and ModularRDF Data on the Web数据来源：http:/esw.w3.org/topic/SweoIG/TaskForces/CommunityProjects/LinkingOpenData
