PowerPoint PresentationSlide 2Slide 3Slide 4Slide 5Slide 6Slide 7Slide 8Slide 9Slide 10Slide 11Slide 12Slide 13Slide 14Slide 15Slide 16Slide 17Slide 18Slide 19Slide 20Slide 21Slide 22Slide 23Slide 24Slide 25Slide 26Slide 27Slide 28Slide 29Slide 30Slide 31Slide 32Slide 33Slide 34Slide 35Slide 36Slide 37Slide 38Slide 39Slide 40Slide 41Slide 42Slide 43Slide 44Slide 45Slide 46Freely Available String Similarity PackagesSyntactic SimilaritySemantic SimilarityName Matching and WordNetExample of entry in WordNetExample Graph Segment in WordNetElements of WordNetWordNet based similarity measuresCosynonymyExampleQuestions to think aboutGloss OverlapGloss Preprocessing ConsiderationsGloss PreprocessingProperty MatchingJaccard SimilarityShortcomings of Jaccard SimCosine SimilarityUnit Circle RefresherFormulaSlide 67Matching text by their entitiesEntity Extraction Using Stanford NERBackground on Stanford NERUsing Stanford NER AppropriatelyExample CodeExample Code (cont)Combining Multiple CriteriaSemantic Sim Score FormulaSlide 76Computing ontology similaritySlide 78Semantic Similarity Matrix1Ontology Alignment and Semantic SimilaritySlides created by Jeff Partyka2Lecture Outline1. Aligning ontologies in Jena2. Ontology Alignment and Semantic Simiarity3. Name based matching methods4. WordNet based matching methods5. Property matching methods using bag-of-word approaches6. Entity Extraction7. Combining ResultsWhat is ontology alignment?•Ontology Alignment is a mapping between the concepts, relationships and instances of 2 or more ontologiesOnt AOnt BWhy do we care?•Due to the diversity of ontologies and other information on the Semantic Web, every person/organization will build an ontology according to their own philosophy•Yet there are many real world use cases for finding the semantic similarity between related ontologies (aligning data models)Why do we care? (cont)•In other situations, we may know of the existence of certain ontologies, but do not know how similar they are. Thus, we don’t know if the ontologies should be aligned to begin withExampleExampleOne Possible AlignmentQuestions about the Alignment•In the previous slide, the ontologies and their proposed alignment can be questioned in many ways•In ontology O1, should ‘Residential Area’ be a subclass of the concept ‘Roads and Ferries’?•How similar are the O1 concept ‘Residential Area’ and the O2 concept ‘Address Area’?•What about Traffic Area (O1) and Enclosed Traffic Area (O2)?Questions about the Alignment (cont)•How similar is the property Road.Length in O1 to the property Road.Miles in O2? Is Road.Length measured in miles?•What about the similarity between Junction in O1 and Intersection in O2? How much should their relationships determine their similarity?•Do these concepts have instance information that can explain their semantics?A Few of Types of Dissimilarity•Same meaning, different syntax: The concepts mean the same thing, but are named differently (Traffic Area, Enclosed Traffic Area). They are semantically similar, but syntactically different•Same syntax, different meaning: Ferry in O1 and Ferry O2 have the same name. But given their properties, they may mean slightly different things. They are semantically dissimilar but syntactically similarA Few of Types of Dissimilarity (cont)•Relationship dissimilarity: The dissimilarity between Junction in O1 and Intersection in O2 is partly based on their respective relationships •Property dissimilarity: Properties may have different names and instances with values for that property•Also, Road.Length in O1 and Road.Miles in O2 seem similar, but may use different units to measure lengthA Few of Types of Dissimilarity (cont)•Instance dissimilarity: concepts may have the same meaning and similar relationships, but the set of instances they each have may imply different semanticsManual Ontology Alignment•We can use the constructs owl:equivalentProperty, rdfs:subclassOf, owl:sameAs, owl:differentFrom, etc., to explicitly define relationships between concepts, properties or individuals in Jena:Jena ExampleAutomated Alignment?•Aligning ontologies manually is fine if the ontologies are well understood •But because of the changing, prolific nature of the Semantic Web, many new ontologies are being created every day•We need an ontology alignment process that is as automated as possible. Is this feasible? What needs to be done to realize this?What it means to align ontologies•Aligning ontologies can take on many different meanings•There is no 1 way to do it, and it is often driven by the use case•Most commonly, ontology alignment means comparing all of the respective concepts between two ontologies, where a similarity score between 1 concept from each ontology is generatedSemantic Similarity•Automated alignment of ontologies via concept matching means that we need a measure to provide a similarity score•Semantic similarity measures frequently report their scores in the interval with 0 indicating no similarity and 1 indicating that the concepts are identical•[0,1], •There are many, many semantic similarity algorithmsScore Thresholds•For simplicity, when determining if two concepts are similar, we use a threshold score to help us•When scoring similarity in the interval [0,1], a threshold score of .6 is often used•Any semantic similarity score between 2 concepts, 1 from each ontology, that is > =.6 indicates that the concepts match. If the score is < .6, that indicates that the concepts do not matchSemantic Distance•Another method used to measure the similarity between 2 concepts is to calculate their semantic distance•The values for semantic distance are often measured in the interval [0,1]•Semantic Distance = 1 – Semantic Similarity•Concept Matching ApproachesMatchings may be generated in several ways – some approaches are:(1: Name Matching(2: Structure Matching(3: Instance MatchingEmail emailAddressCounty DSPKitsap KingstonWahkiak Puget IslandCOUNTYNAME CIDTRAIL RANGE DR 96KITSAP 97?Name Matching•Concepts are matched by measuring the similarity between their names•This can be done using (1: string matching methods (2: meaning matching methods•String matching methods are syntactic, whereas meaning matching methods are semanticString-based name matching methods•We will discuss the following topics in string based name matching of concepts:- Normalization- Hamming Distance- N-grams- Edit