Upload
adele-price
View
213
Download
0
Embed Size (px)
Citation preview
1
SOCoP 2012 Workshop:
Vision and Strategy
Gary Berg-CrossSOCoP Executive Secretary
Nov. 29-30, 2012
U. S. Geological Survey National Center 12201 Sunrise Valley Dr., Reston VA
2
Outline for a VoCamp-Style Workshop
VoCamp Style is informal Workshop Ingredients
1. Goals
2. Workgroup Teams
3. Logic of Phased Structure Sessions
4. Lightweight Methods
5. From Conceptualizations to Formalizations
6. Cautions and Tradeoffs
7. Ontologies and Ontology Patterns
We are not here to teach a new discipline, but to employ multiple disciplinesas a team.
3
Goals: “Group Work on Semantic Models of Interest”
Building on past workshops and interests: We seek clarified agreement & reduced
ambiguities/conflicts on geospatial/earth science phenomena that can be formally represented in:1. Constrained, engineered models to support understanding,
reasoning & data interoperability and/or2. Creation of general patterns that provide a common
framework to generate ontologies that are consistent and can support interoperability.
We like data-grounded work since: Much of the utility of geospatial ontologies will likely
come from their ability to relate geospatial data to other kinds of information.
4
Workgroups Include Multiple Roles:Semantic Engineering is a Social Process
Ontologist
Domain/Data Expert
Facilitator
Note taker
5
Logic of Work Sessions
Start Group organization & Introductions, goals and process
After lunch Group Work on Concepts, Vocabulary & Model(s)
After break Group Work on Draft Models
At end of day Group Reports on status
At end of day Report back to whole and wrap up
2nd day draft final model & initial formalizations
After break Work Groups polish, formalize models
After lunch firm up products and test against data
After break Prepare Report
Day One Intro, Topics, Methods..
6
Lightweight Methods & Products
Choose lightweight approaches grounded by scenarios and application needs. Low hanging fruit leverages initial vocabularies and existing
conceptual models to ensure that a semantics-driven infrastructure is available for use in early stages of work
Reduced entry barrier for domain scientists to contribute data
Simple parts/patterns & direct relations to data
Ecological..
Triple like parts
Constrained not totallySpecified.Grounded
7
How Ontologies Arise from “Analysis and Conceptualization”
Adapted liberally from Guarino’s 1998Formal Ontology in Information Systems
1 World Situations
Perceived Distinctions& Situated Experience
.. Stream Reaches are segments of surface water with similar hydrologic
characteristics
Conceptualizationstarts to adaptively model
(part of) the world
2 Abstraction
Stream Reach distinctionRationalized Intuition expressed in
Some conceptual language
Things D in world state W with conceptual relations R
C= <SW, H, S>SW =surface water
H=has S = Segments
Pragmatic validation
Real World Semantics
Domain folks and Modelers
8
Ontological Engineering – Formalization
1 World Situations
Interaction..That is a
type of Stream
segment
C Conceptualization
model (part of) the worldSay D Hydrology
In some Conceptual Domain space
UML OWL
3. To express C weneed Language L (Term StreamX corresponds to entity in world)
We assign interpretative Functions I &Commitment K. K=<C,I)
Models forDomain DExpressibleIn L
IntendedModelFitting
COntologyModels forDExpressedIn L using K
Adapted liberally from Guarino’s 1998Formal Ontology in Information Systems
Models defines relationship between L syntax and interpretations
Logical ModelLogical ModelTheories are Theories are consistent sets of sentences; consistent sets of sentences; closed under logical closed under logical consequenceconsequence
9
Caution(s)
No single (general or domain) taxonomy or ontology could satisfy all needs. That’s part of the reason we have so many alternative ones.
We can make quality, useful ones for one or more purposes. We have intended models of focus. But if we make commitments that make strong claims that are
greater our understanding and focus we may be making a mistake(s).
Smith and Mark describe a good/quality ontology as one that is: “neutral”,
And/or perhaps explicit about what it is not neutral about computationally tractable, acceptable to and reusable by domain people
One might add that it is rationalized.
10
Many Challenges – more on this later….
These include tradeoffs between generality and precision. Tom Bittner, for example, notes:
“The need for geometric representations, many of which rely on relatively precise boundaries, conflicts with the need for sophisticated classification systems for scientific and integration purposes. “
Vagueness and the tradeoff between the classification and delineation of geographic regions - an ontological analysis
http://www.geoinfo.info/geoinfo2012/programa.php
But these, of course, make the work interesting…..
11
Ontologies and Ontology Patterns
Many levels & types of ontologies Something like DOLCE is quite complex and so are
domain ontologies like SWEET One may pick some small repeating patterns (ODPs) out
of large ontologies. ODPs, like OWL, are tools for ontologies
They are more easily understandable with good explicit documentation for design rationales
RobustCan be used to build on modularly for reoccurring problems needing representation
Capture best practicesShould help bridging/integrating ontologies
We focus on content ODPs using domain expertise rather than logical ODPs etc.owl:Class:_:x rdfs:subClassOf owl:Restric(on:_:y
Inflammation>on rdfs:subClassOf (localizedIn some BodyPart)
11
Dolce UltraLite?
Apps( Uses Cases &
Demos)
“Geo”-Domain
“Geo” Top
Patterns
12
Ontology-ODP RelationsSmall DUL Portion <owl:Class rdf:ID="SocialObjectAttribute"> <rdfs:label xml:lang="en">Social attribute</rdfs:label> <rdfs:subClassOf> <owl:Restriction> <owl:onProperty> <owl:ObjectProperty rdf:ID="isRegionFor"/> </owl:onProperty> <owl:allValuesFrom> <owl:Class rdf:ID="SocialObject"/> </owl:allValuesFrom> </owl:Restriction> </rdfs:subClassOf> <rdfs:label xml:lang="it">Caratteristica sociale</rdfs:label> <rdfs:subClassOf> <owl:Class rdf:ID="Region"/> </rdfs:subClassOf> <rdfs:comment>Any Region in a dimensional space that is
used to represent some characteristic of a SocialObject, e.g. judgment values, social scalars, statistical attributes over a collection of entities, etc.</rdfs:comment>
</owl:Class> <owl:Class rdf:ID="WorkflowExecution">…
ODP
Abstract From
Use for
NewOntology Design
“Unfriendly logical structures, some large, hardly comprehensibleOntologies” (Aldo Gangemi)
13
ODP Rationale –Reuse, Minimal Constraints..
Problem It is hard to reuse only the “useful pieces" of a comprehensive
(foundational) ontology, and the cost of reuse may be higher than developing a scoped
ontology for particular purpose from scratch “For solving semantic problems, it may be more productive
to agree on minimal requirements imposed on .. Notion(s) Werner Kuhn (Semantic Engineering, 2009)
Solution Approach Use small, well engineered, modular starter set ontologies
with explicit documentation of design rationales, and best reengineering practices
These serve as an initial constraining network of “concepts” with vocabulary which people may build on/from for various purposes.