1. Semantic metadata in the Catalogue Frédéric Houbie

2. Overview Context and motivation New catalogue semantic functionalities Storing ontologies Tagging metadata Semantic queries Questions, issues & perspectives

3. Context and motivation ERDAS has been developing semantic enhancements to the CS/W catalogue, in the scope of several projects SWING : FP6 European research project dealing with semantic discovery and chaining of GIS services (http://www.swing- project.org)http://www.swing- project.org These projects involve the semantic tagging and discovery of services (and other resources), in order to achieve service chaining This has lead to the application of semantic techniques to services cataloguing

4. Problems to solve Querying the catalogue is currently limited to string matching tedious because of verbose and complicated queries Ontologies are used to tackle two main problems Structure mapping : query across heterogeneous data structures Concept matching : use knowledge from domain ontologies to improve results

5. Problem 1 : Structure mapping We have various MD formats and data structures (ISO19115, ISO19119, OWS Capabilities, EO, sensorML, FGDC, proprietary…) Solution ? Choose a unique metadata structure ontology –Re-use ISO19115 ? –a simpler one ? (eg. a flattened subset of ISO19115) Store MD in their original structure, but tag the elements with the structure ontology This way : query transparently services/resources across the various MD structures

6. Problem 2 : use Domain Ontologies to enrich metadata Tag existing metadata with domain ontologies, and leverage existing knowledge from those ontologies in the query Example –find all Raster services (i.e. matching subconcepts of formats#Raster) –Find all services that provide data that is relevant to air navigation If metadata are tagged with concepts from widely accepted domain ontologies (maybe not specific to GIS), any new domain knowledge can be instantly leveraged

7. Required changes What is needed to achieve this ? Store ontologies Link existing metadata to stored ontologies Query the catalogue using the semantic objects (concepts, properties, …)

8. OASIS Recommendation There is a draft recommendation paper at the OASIS regarding an “ebXML registry profile for OWL” This paper details how to model ontologies within an ebXML registry define stored procedures at the DB level to perform semantic queries This recommendation is the basis for the work described here

9. Registering Ontologies Using the data model defined in the OASIS recommendation This model uses ebRIM specificities to model OWL constructs Add support in the Catalogue to harvest OWL files So far : only a subset of OWL-lite constructs are supported Ontology, Class, SubClass, Property, SubProperty

10. Registering Ontologies Sample ‘formats’ ontology, and its ebRIM representation : <rdf:RDF xmlns="http://example.com/ontologies/gis_datatypes">

11. Tagging metadata Once ontologies are stored within the catalogue, one can easily link existing MD registry objects to ontology classes or properties This is independent of the actual data model of the MD objects, and can be applied to already existing ebRIM records, without altering their structure One MD registry object can be classified by several ontologies representing different domains

12. Tagging metadata

13. Querying semantic resources There is a need for new operators in OGC filters The OASIS recommendation defines many DB stored procedures to implement semantic operators Issues concerning these operators How to introduce them inside OGC filters ? Re-use existing semantic query language? (SPARQL ?) What are exactly the operators needed ? Does this fit within the OGC Filter specification ? In the current prototype : Concept(URI) : identifies a Concept object from an URI isInstanceOf(obj, class) : boolean operator; true if ‘obj’ is classified by ‘class’ or any of its subclasses satisfiesLink(obj1, linkClass, obj2) : true if ‘obj1’ has an association of type ‘linkClass’, or any of its subclasses, towards ‘obj2’ More to come (support for equivalency, transitivity, …) Copyright © 2007, Open Geospatial Consortium, Inc., All Rights Reserved.

14. Querying semantic resources This query will match the ‘GML’ concept, and any of its subconcepts The GML concept is specified using its URI, as in OWL /RegistryObject/@id http://example.com/ontologies/gis_datatypes#GML

15. Querying semantic resources (2)

16. What ontologies are needed ? Structure ontologies Define representation models Example : ISO19115 ontology Not ‘true’ semantics; more a mapping definition Domain ontologies Define reality Example : ontology of output formats, of countries, of ISO categories Used to classify MD elements (that belong to some representation model)

17. Examples : WFS & SLD We have SLD rules, with their use described with an ontology of feature types (using SemanticIdentifier attribute) We want to be able to catalogue those rules, then be able to query the catalogue to find the most suitable rules for a given WFS layer, assuming this layer has also been tagged with the same ontology

18. How to use ontologies to solve that WFS FT1 FT2 FT3 FeatureTypes Roads Hydro Building Highways Path Lake Sea River … … SLD Rule give me all the rules that can portray WFS feature type X

19. Another use case These mechanisms allow for immediate use of new domain knowledge You have various resources in a catalogue, with data services and SLD rules classified with concepts such as ‘GML’, or ‘Airport’ For a specific use case, domain experts issue a new ontology that states that for an aircraft mission planning, one needs –vector data about airfields –raster data about meteorology –vector data to be portrayed with enough precision for scale 1:10000 Then, inserting that ontology into the catalogue, without modifying existing MD, immediately allows requests like : –Find all resources suitable for an aircraft mission planning on BBOX …

20. Questions, Issues & Perspectives In terms of OGC specifications, this can lead to OWL ebRIM Extension Package OGC Filters extension Input formats for ontologies So far : OWL Extend to other ontology languages ? Add MD Output formats RDF : to have metadata available to outside reasoners Plug reasoner in CS/W ? How far should we go with reasoning inside CS/W ? Is concept matching enough ? Too much already ? At this stage : very simple concept matching. What about mediation of data structures ?

21. Questions, Issues & Perspectives (2) Query language ? Use SPARQL ? Maybe enhancing the Filter spec would profit also other services (WFS ?) More semantic operators will be implemented (support OWL equivalency, transitivity, …) CS/W should not be limited to cataloguing services Portraying rules, datasets, feature type structures… Domain ontologies can be used across resource types to link them semantically –f.i. : the ‘highway’ concept can be used to tag a WFS service and also an SLD rule that can portray highways