1. 1 Ontology Enabled Data Discovery and Integration Kai Lin San Diego Supercomputer Center University of California, San Diego A. K. Sinha, Z. Malik, A. Rezgui, A. Dalton Virginia Tech

2. 2 Motivations A better way to discover and understand datasets Use the knowledge in ontologies to find datasets A better way to query datasets Query through ontologies without knowing the schemas A better way to integrate multiple datasets Integrate multiple datasets on-the-fly if they are mapped to ontologies

3. 3 What Is Ontology A formal, explicit specification of a shared conceptualization unambiguous definition of all concepts, attributes and relationships machine-readability commonly accepted understanding conceptual model of a domain

4. 4 Why Represent Domain Knowledge as Ontology Separate domain knowledge module from the operational module Configurable knowledge module Share and reuse domain knowledge Analyze domain knowledge

5. 5 What’s Inside An Ontology? Concepts: Classes + Class-hierarchy –instances Properties: often also called “Roles” or “Slots” –labeled instance-value-pairs Axioms/Relations: –relations between classes (disjoint, covers) –inheritance (multiple? defaults?) –restrictions on slots (type, cardinality) –Characteristics of slots (symm., trans., …) reasoning tasks: –Classification: Which classes does an instance belong to? –Subsumption: Does a class subsume another one? –Consistency checking: Is there a contradiction in my axioms/instances?

6. 6 Resource Description Framework (RDF) XML Schema is not enough for semantics only describe Grammar, i.e. syntax of single documents can not express inheritance for concepts no means to express complex integrity constraints in an unambiguous way Resource Description Framework (RDF) an infrastructure for the encoding, exchange and reuse of structured metadata Peter Morris Peter Morris page.html The author of ‘page.html‘ is Peter Morris What is the “correct” way of expressing it?

7. 7 RDF Idea RDF is intended to provide a simple way for making statements about resources Resources objects that are uniquely identified by an URI (Uniform Resource Identifier) Anything can have a URI. an entire Web page, a whole collection of pages e.g. an entire Website, object that is not directly accessible via the Web such as a printed book. Property a specific aspect, characteristic, attribute, or relation used to describe a resource has a specific meaning, defines its permitted values Lives-In, CarColor, WorkFor, HasA, IncludedIn, hasAuthor… Statement a specific resource together with a named property plus the value of that property for that resource. Each RDF statement can be written down as a triple (Subject, Property, Object) or a graph Resource property Value Resource

8. 8 A RDF Example <rdf:RDF xmlns:rdf = “http://www.w3.org/1999/02/22-rdf-syntax-ns#” xmlns:dc = “http://purl.org/dc/elements/1.1/”> Peter Morris http://www.polleres.net/page.html http://www.polleres.net/peter Peter Morris http://purl.org/dc/elements/1.1/creator hasName April 1,2004 creationDate English http://purl.org/dc/elements/1.1/language

9. 9 A General RDF Format value of property-A value of property-B Value-C Convention: A capital letter to start a type (class) name A lowercase letter to start a property name

10. 10 RDF Schema (RDFS) Core Class rdfs:Resource rdfs:Literal rdf:XMLLiteral rdfs:Class rdfs:Property rdfs:DataType rdfs:Container Core Property rdf:type rdfs:subClassOf rdfs:subPropertyOf rdfs:domain rdfs:range rdfs:label rdfs:comment RDFS is a simple ontology language RDF: triples for making assertions about resources RDFS extends RDF with “schema vocabulary”, e.g.: –Class, Property –type, subClassOf, subPropertyOf –range, domain  representing simple assertions, taxonomy + typing

11. 11 RDFS Example ResourceClass Property HoverVehicle Company Number Vehicle SeaVehicleLandVehicle subClassOf type producedBy type numberOfEngine

12. 12 RDFS too weak to describe resources in sufficient detail: –No localised range and domain constraints Can’t say that the range of hasChild is person when applied to persons and elephant when applied to elephants –No existence/cardinality constraints Can’t say that all instances of person have a mother that is also a person, or that persons have exactly 2 parents –No transitive, inverse or symmetrical properties Can’t say that isPartOf is a transitive property, that hasPart is the inverse of isPartOf or that touches is symmetrical –No in/equality Can’t say that a class/instance is the same as some other class/instance, can’t say that some classes/instances are definitely disjoint/different. –No boolean algebra Can’t say that that one class is the union, intersection, complement of other classes, etc. Limitations of RDFS

13. 13 OWL Language - Overview Three species of OWL –OWL DL stays in Description Logic fragment –OWL Lite is “easier to implement” subset of OWL DL –OWL Full is union of OWL syntax and RDF OWL DL based on Description Logic –In fact it is equivalent to SHOIN (D n ) DL OWL DL Benefits from many years of DL research –Well defined semantics –Formal properties well understood (complexity, decidability) –Known reasoning algorithms –Implemented systems (highly optimised) OWL full has all that and all the possibilities of RDF/RDFS which destroy decidability Full DL Lite

14. 14 Full DL Lite OWL Full Allow meta-classes etc OWL DL Negation (disjointWith, complementOf) unionOf Full Cardinality Enumerated types (oneOf) OWL Light (sub)classes, individuals (sub)properties, domain, range intersection (in)equality cardinality 0/1 datatypes inverse, transitive, symmetric hasValue someValuesFrom allValuesFrom RDF Schema OWL Layers (Lite, DL, Full)

15. 15 Ontology Inconsistency You may define Classes were no individual can fulfill its definition. Via reasoning engines such a definition can be found also in big ontologies. –Cow ≡ Animal ⊓ Vegetarian –Sheep ⊑ Animal –Vegetarian ≡  eats  Animal –MadCow ≡ Cow ⊓  eats.Sheep

16. 16 Open/Close World Assumption Close World Assumption –The fact in the ontology describe completely what I know, all that is not in the ontology is assumed to be false.. Open World Assumption (used in OWL) –There are something not described by the ontology An ontology says: There is a train at 14:00 There is a train at 15:00 Is there a train at 17:00? no by Close World Assumption unknown by Open World Assumption

17. 17 Resource Discovery in GEON A Resource Registration System for Data Providers –Register ontologies (domain knowledge) –Register datasets with metadata including data access information –Optionally register datasets to ontologies (which is crucial for data integration and smart search) A Search Engine for Data Users –Metadata based search –Spatial coverage based search –Temporal coverage based search –Concept based search Both are available through a public portal on the web

18. 18 Metadata (ADN) GEON Data Registration System Resource Registration System SRB Metadata (ADN) Metadata (ADN) Metadata (ADN) Excel GeoTIFF Shapefile Catalog General InformationOntology Annotations Access Control Subjects Format Keywords Spatial coverage's Temporal coverage's ………… Integrated Resources Log Resource Metadata GEON Search Resource Schemas

19. 19 Database Registration Table View Original Database Table Def View Def Published Database select tables and views to register GEON Mediator GEON JDBC Driver Application

20. 20 Write Protection Mediator Database UPDATE B Only accepts SELECT statements Rejects any requests other than SELECT A B C B

21. 21 Read Protection on Unregistered Tables and Views Mediator Database SELECT * FROM A An unregistered table or view is invisible to an end user The data in the table can’t be viewed by SELECT statement The schema can’t be fetched A B C B

22. 22 Item Level Ontological Data Registration for Discovering The search engine uses ontologies to find more results, for example, the fact that Polygon is a subclass of GeometricalObject is used in the searching. Rectangle CirclePolygonSurface GeometricalObject_2D Ontology: Dataset Properties mentions uses has instances Search for GeometricalObject_2DReturn datasets associated with Polygon

23. 23 Data Integration Challenges: Heterogeneities Syntactical Heterogeneity heterogeneous data format e.g. 02-04-2004 vs. 02/04/04 Structural Heterogeneity heterogeneous data models and schemas e.g. 02-04-2004 is saved as three columns or one columns Semantics Heterogeneity fuzzy metadata, terminology, “hidden” semantics, implicit assumptions GEON Preferred Solution: Datasets are semantically registered first Heterogeneities is resolved by registration

24. 24 Database Integration Integration at three levels Level 1: Federation Based Integration Users should be knowledgeable to each databases Level 2: View Based Integration The intended users are somebody who want to do integration for others or make integration results reusable Level 3: Ontology Based Integration The easiest way for end users

25. 25 Level 1: Federation Based Integration C AB G D F E C AB D GF E Mediator backend SELECT * FROM A, E WHERE …… Use SQL to query the federated database Structural and semantic heterogeneity should be solved by users themselves

26. 26 Level 2: View Based Integration C AB G D F E C AB D GF E Mediator backend SELECT * FROM V, W WHERE …… Allow defining views on top of the federated databases Allow hiding the original backend schemas Integration results can be shared and reused VW

27. 27 Level 3: Ontology Based Integration Require ontology annotations for backend databases Use simple ontology query language to query the integrated database Users don’t need know the backend schemas and local semantics C AB G D F E C AB D GF E Mediator backend Ontology Based Query

28. 28 Ontology Enabled Data Integration Ontology Enabled Semantic Integration Challenges for Computer Scientists and Domain Scientists –Computer Scientists: build an integration system based on the ontological registration of datasets –Domain Scientists: create domain ontologies –Data Providers: register datasets to ontologies Ontology1 Ontology2 ontology3 dataset1dataset2dataset3 dataset4

29. 29 Ontological Data Registration for Data integration Registering a dataset to an ontology for data integration is a procedure to generate a partial model of the ontology from the dataset itself From registration dataset individualsontology p Not all the constraints in the ontology are satisfied by the generated individuals

30. 30 Associate one or more columns under an optional SQL condition to a selected class in the ontology Provide a mapping method if no explicit names of individuals should be generated Registering Relational Tables to Ontology Classes ……Latitude……Longitude…… 23.547.9 …… Location (23.5, 47.9) is the name of an individual of the class Location Same name indicates the same location RockSample GeologicAge …… Jurassic/Triassic Precambrian ………… GeologicalAge PrecambrianCenozoicPaleozoic

31. 31 Registering Tables to Ontology Object Properties Associate two entities which are already registered to the domain class and the range class of a selected object property in the ontology ……RockSampleID……PERIOD…… Rock GeologicAge hasAge

32. 32 ODAL (Ontological Database Annotation Language) <odal:NamedIndividuals odal:id="RockSample" odal:database="VTDatabase"> Samples RockTexture RockGeoChemistry ModalData MineralChemistry Images ssID GUI generate to ODAL processor The values in the column ssID of the table Samples, RockTexture, RockGeoChemistry, ModalData,MineralChemistry and Images represent instances of RockSample Create a partial model of ontologies from database Independent on any GUI Independent on any concrete implementations reusable

33. 33 ODAL: Import Ontologies The Ontologies used for annotating a database can be imported as follows: <odal:ODAL xmlns:rdf = “http://www.w3.org/1999/02/22-rdf-syntax-ns#” xmlns:owl="http://www.w3.org/2002/07/owl#" xmlns:odal = “http://www.sdsc.edu/odal#” > ……

34. 34 ODAL: Database Connection Declaration The target databases for making annotation is declared as follows: <odal:ODAL xmlns:rdf = “http://www.w3.org/1999/02/22-rdf-syntax-ns#” xmlns:owl="http://www.w3.org/2002/07/owl#" xmlns:odal = “http://www.sdsc.edu/odal#” > …… Oracle 9.1.21 oracle.sdsc.edu 3456 Publications ……

35. 35 ODAL: Simple Named Individuals <odal:NamedIndividuals odal:id="BookInTableBookPrice" odal:database="PublicationDatabase" > Collections book-price ISBN Suppose the book ontology contains a class Book and the schema Collection contains a table book-price with a column ISBN. odal:id gives a name to the declaration, and represents the set of the individuals generated by the statement. The statement says that each value in the column ISBN represents a book individual.

36. 36 ODAL: Named Individuals from Multiple Columns California Rock-Sample Latitude Longitude Suppose an ontology contains a class Location and a database table Rock-Sample with two columns Latitude and Longitude. The statement says that a pair of latitude and longitude gives a location

37. 37 ODAL: Named Individuals with Conditions employee EmployeeId ]] employee EmployeeId ]] A condition in an odal:Condition element should be a boolean expression which is valid to be used in any WHERE clauses of SQL queries

38. 38 ODAL: Data Type Property Declaration Person ssn person …8…1234-56-7890… …age…SSN… Person double hasAge

39. 39 Usually we don’t make join on individuals cross different resources A set of datatype properties can be declared as a key for a class in the ontology. We do join cross multiple resources based on keys. e.g. { hasLatitude, hasLongitude} can be declared as a key of Location Two locations from different resources are same if they have the same latitude and longitude Conditions for Joining from Different Resources Rock RockSampleID 10001 …... RockID 10001 …… We don’t know whether 10001 represents the same rock in the two resources. By default, we assume they are not.

40. 40 SOQL (Simple Ontology Query Language) Query single or integrated resources via ontologies (i.e., high level logical views) independent on any physical presentation (i.e. schemas) RockSampleLocation ValueWithUnit float location hasSiO2 value latlong unit string SELECT X.location.*; FROM RockSample X WHERE X.location.lat > 60 AND X.location.long > 100 AND X.hasSiO2.value < 30 AND X.hasSiO2.unit =‘weightPercetage ’ GUI generate to SOQL processor

41. 41 The Architecture of GEON Semantic Mediator Portal or Application Mediator JDBC Driver GUI SOQL Semantic Query Rewriter SOQL Parser Ontology Reasoner SOQL Processor Spatial SQL against federal schemas SQL Parser OWLODAL Query Execution Query Optimization Query Planning Internal Database OracleDB2MySQL SQL Server PostgreSQL PostGIS ODAL Processor

42. 42 SELECT X.code, X.location.* FROM SeismicStation X, Railroad Y WHERE distance(X.location, Y.geometry) < 1 SELECT X2.stationcode, X2.lat, X2.lon FROM railroads_of_the_united_states X1, stationdatatable X2 WHERE distance(X1.the_geom, MakePoint(X2.lat, X2.lon)) < 1 GEON SOQL GUI SOQL Processor Railroad shapefile Seismic Stations Schema Mediator distance(X1.the_geom, MakePoint(X2.lat, X2.lon)) < 1 SELECT X1.the_geom FROM railroads X1 Question: Finding all seismic stations within 1 mile from railroads SELECT X2.stationcode, X2.lat, X2.lon FROM stationdatatable X2 WHERE bounding box condition

43. 43 Questions?

44. 44 How to Connect to GEON Databases Download GEON JDBC Driver Use the following code to create a connection // load driver Class.forName ("org.geongrid.jdbc.driver.Driver"); // set the mediator URL String url = "jdbc:geon://geon01.sdsc.edu:2532/GEON-63cb404c-6038-11d9-a69f”; // open the connection Connection conn = DriverManager.getConnection(url, "geonuser", "geongrid"); GEON JDBC protocol The host name and port number of GEON Mediator GEON ID Note: the original account information is invisible to end users