Ontology 101 - Kendall & McGuiness

+

Ontology 101: An Introduction to Knowledge
Representation, the Web Ontology Language (OWL) &
Ontology Development
Elisa Kendall, Thematix & Deborah McGuinness, RPI / McGuinness Associates
6 June 2011

+ Content management
Mars was photographed by the Hubble Space Telescope in August 2003 as 2 2
the planet passed closer to Earth than it had in nearly 60,000 years.
Image Credit: NASA, J. Bell (Cornell U.) and M. Wolff (SSI)
A sunset on Mars creates a glow due to the
presence of tiny dust particles in the
atmosphere. This photo is a combination of
four images taken by Mars Pathfinder, which
landed on Mars in 1997. Image credit:
NASA/JPL

Recent images from instruments on board the
Mars Reconnaissance Orbiter take much more
detailed, narrower views of specific features of the
Martian surface. Image credit: NASA/JPL

The Planetary Data Store (PDS) is a distributed repository of 40+ years’ imagery & data
taken by a range of instruments on many diverse missions, available for scientific
research.
Copyright © 2011 Thematix & McGuinness Associates

+ Smart search
3 3

Provenance/sources for tracking family members in the 19th
century include early census data (often error prone), military
records, passenger & immigration lists, online documents (e.g.,
county histories, church histories, etc.)

• Historical/forensic research requires cross-domain search of a wide variety of resources
within a given geo-spatial/temporal context
• Similar capabilities are essential for business intelligence, law enforcement, government
applications – all require terminology reconciliation


+ Merchandising
4 4
Attribute 1
Attribute 2
Leg Room
Attribute 4
Attribute 5
Attribute 5
On-Off Access
“Comfort- Attribute 7
Attribute 8
Oriented” Attribute 9
Flight Media Options
Attribute 11
Attribute 12
Meal Options
Attribute 14
Attribute 15
Time of Day
Attribute 17
Attribute 18
Distance to Gate
Attribute 20
Attribute 21
Flight Characteristics
Attribute N
…


+ Search Engine Optimization
5 5

 Use search engines as a front door to transaction

 Dramatically increase relevance by inclusion of more
meaningful, synonymic tags

 Enrich search results, adding ratings, video, prices,
avails, amenities, locality etc.


+ Tutorial Outline
6 6

 Introduction to Knowledge Representation
 OWL Basics
 Tools & Applications


+ Part 1: Introduction to Knowledge Representation
7 7

 A little background and a few definitions
 Layers of abstraction & conceptual modeling
 Classifying ontologies
 A little methodology


+ Definitions
8 8
 An ontology is a specification of a conceptualization.
– Tom Gruber
 Knowledge engineering is the application of logic and ontology to the task of
building computable models of some domain for some purpose. – John Sowa
 Artificial Intelligence can be viewed as the study of intelligent behavior
achieved through computational means. Knowledge Representation then is
the part of AI that is concerned with how an agent uses what it knows in
deciding what to do. – Brachman and Levesque, KR&R
 Knowledge representation means that knowledge is formalized in a symbolic
form, that is, to find a symbolic expression that can be interpreted. – Klein and
Methlie
 The task of classifying all the words of language, or what's the same thing, all
the ideas that seek expression, is the most stupendous of logical tasks.
Anybody but the most accomplished logician must break down in it utterly;
and even for the strongest man, it is the severest possible tax on the logical
equipment and faculty. – Charles Sanders Peirce, letter to editor B. E. Smith of
the Century Dictionary


+ What is an Ontology?
An ontology specifies a rich description of the
9 9
 Terminology, concepts, nomenclature

 Properties explicitly defining concepts

 Relations among concepts (hierarchical and lattice)

 Rules distinguishing concepts, refining definitions and relations (constraints,
restrictions, regular expressions)

relevant to a particular domain or area of interest.


+ Logic and Ontology
10 10
 Predicate logic is harder to read than the original English, but is
more precise:
Every semi-trailer truck has at least 3 axles.

(∀ x)(((SemiTrailerTruck(x) ∧ (∃ y)(SemiTrailer(y) ∧ (hasPart (x,y))) ∧
(SemiTrailerTruck(x) ∧ (∃ z)(TractorUnit(z) ∧ (hasPart (x,z))))
⊃ (∃ s)(set(s) ∧ (count(s,(≥3))
∧ (∀ w)(member(w,s) ⊃ (Axle(w) ∧ hasPart(x,w))) )).

 Logic is a simple language with few basic symbols.
 The level of detail depends on the choice of predicates – these
predicates represent an ontology of the relevant concepts in the
domain.
 Different choices of predicates represent different ontological
commitments.
* Derived from Knowledge Representation: Logical, Philosophical, and Computational Foundations,
John F. Sowa, Brooks/Cole, Pacific Grove, CA, 2000.


+ Ontology-based Technologies
11 11
 Ontologies provide a common vocabulary for use by independently
developed resources, processes, services
 Agreements among organizations sharing common services can be
made with regard to their usage; the meaning of relevant concepts can
be expressed unambiguously
 By composing / mapping ontologies and mediating terminology across
participating events, resources and services, independently-developed
services can work together to share information and processes
consistently, accurately, and completely
 Ontologies also ensure
 Valid conversations among agents to collect, process, fuse, and
exchange information
 Accurate searching by ensuring context using concept definitions
and relations instead of/in addition to statistical relevance of
keywords

+ Features of KR languages
 Vocabulary – a collection of symbols 12 12
 Domain-independent logical symbols (e.g., ∀ or ⊃)
 Domain-dependent constants, identifying individuals, properties, or
relations in the application domain or universe of discourse
 Variables, whose range is governed by quantifiers
 Punctuation that separates or groups other symbols
 Syntax –
 formation rules that determine how symbols can be combined in well-
formed expressions
 rules may be stated in a linear grammar, graph grammar, or independent
abstract syntax
 Semantics –
 a theory of reference that determines how the constants and variables
are associated with things in the universe of discourse
 a theory of truth that distinguishes true statements from false statements
 Rules of Inference –
 rules that determine how one pattern can be inferred from another
 if the logic is sound, the rules of inference must preserve truth as
determined by the semantics

+ Logic
Logics vary from classical FOL along six dimensions: 13 13
 Syntax
 Subsets limit permissible operators or combinations, e.g., propositional logic
(without quantifiers), Horn-clause (excludes disjunctions in conclusions, such as
Prolog), terminological or definitional logics (containing additional restrictions,
e.g., description logics)
 Proof Theory restricts or extends permissible proofs:
• Intuitionistic logic and relevance logic rule out certain extraneous information
• Non-monotonic logics allow introduction of default assumptions
• Access-limited logic restricts the number of times a proposition can be used in a proof;
Linear logic allows a proposition to be used only once
• Modal logic incorporates modal auxiliaries (◊p means p is possibly true; p means p is

necessarily true), temporal logic extends model logic to include always, sometimes
• Intensional logics express concepts such as need, ought, hope, fear, wish, believe, know,
expect, and intend
 Model Theory modifies the truth value of statements in terms of some model of the
world: classical FOL is two-valued; a three-valued logic introduces unknowns;
fuzzy logic uses the same notation as FOL but with an infinite range of certainty
factors (1.0 to 0.0)
 Ontology – frameworks may include support for built-in components, such as set
theory or time
 Meta-language – language for encoding information about objects

+ Description logic
 A family of logic-based Knowledge Representation formalisms 14 14
 Descendants of semantic networks and KL-ONE
 Describe domain in terms of concepts (classes), roles (relationships), and
individuals (instances)
 Distinguished by
 Formal semantics
• Decidable fragments of FOL
• Closely related to Propositional, Modal, and Dynamic Logics
 Provision of inference services
• Sound and complete decision procedures for key problems
• Implemented systems (highly optimized)

 Applications include
 Configuration – product configurators, consistency checking, constraint
propagation, first significant industrial application (e.g., CLASSIC)
 Ontologies – ontology engineering (design, maintenance, integration),
reasoning with ontology-based mark-up, service description and discovery
 Databases – consistency of conceptual schemata, schema integration, query
subsumption (w.r.t. conceptual schemata)


+ Knowledge bases, databases, & ontology
15 15
 An ontology is a conceptual model of some aspect of a particular
universe of discourse (or of a domain of discourse)
 Typically, ontologies contain only “rarified” or “special” individuals,
metadata, representing elemental concepts critical to the domain
 A knowledge base is a persistent repository for
 Ontology & metadata representing individuals, facts, & rules about how they
can be combined or relate to one another
 Metadata, facts & rules only – in some applications and frameworks the
ontology is separately maintained

 Most inference engines require in-memory deductive databases for
efficient reasoning (including commercially available reasoners)
 A knowledge base may be implemented in a physical, external
database, such as a relational database, but reasoning is typically
done on a subset (partition) of that knowledge base in memory


+ Reasoning & Truth Maintenance
16 16
 Reasoning is the mechanism by which the assertions made in an
ontology and related knowledge base are evaluated by an inference
engine.
 In classical logic, the validity of a particular conclusion is retained
even if new information is received.
 This may change if some of the preconditions are actually
hypothetical assumptions invalidated by the new information.
 The same idea applies for arbitrary actions – new information can
make preconditions invalid.
 Generally, there are two issues that a reasoner must address:
 If some conclusion is invalid, which other conclusions are also invalid?
 If some action cannot be performed, which others are at risk?
 The “housekeeping” associated with tracking the threads that support
answering these questions is called truth maintenance.


+ Negation
 If all new information is “positive”, then all prior conclusions should 17 17
remain valid.
 Problems arise if new information negates a prior assumption,
causing it to be withdrawn.
 What does it mean to negate (withdraw) an assumption?
 Conclusive information is not available?
 The assumption cannot be proven?
 The assumption is not provable using certain methods?
 The assumption is not provable given a fixed quantity of time?
 The answers to these questions can result in different definitions of
negation and differing interpretations by non-monotonic reasoners.
 Solutions include chronological and “intelligent” backtracking
algorithms, heuristics, circumscription algorithms, justification or
assumption based retraction, and so forth, depending on the reasoner
and methods used for truth maintenance.
 Reasoning efficiency is dependent, in part, on the algorithms applied
for truth maintenance.


+ Explanations and Proofs
18 18
 When a reasoner draws a particular conclusion, many users and
applications want to understand why?
 Primary motivations include interoperability, reuse, and trust
 Understanding the provenance of the information and results is
crucial, especially when web-based information is involved
 What information sources were used (source)
 How recently they were updated (currency)
 How reliable these sources are (authoritativeness)
 Was the information directly available or derived, and if derived, how
(method of reasoning)
 Methods used to explain why a reasoner reached a particular
conclusion include explanation generation and proof specification


+ Abstraction Layers
• Identify subject areas
Contextual 19 19
Ontology

Conceptual
• Define the meaning of things in the organization

• Describe the logical representation of properties Ontology, Conceptual ER, Conceptual
Logical
Business Process …

• Describe the physical means by which data is stored ER, Relational, XML Schema
Physical

XML, source code, scripting languages,
• Represent the coding language on a specific development platform stored procedures…
Definition

Physical KBs,
• Hold the values of the properties applied to the data in a schema databases, asset repositories…
Instance

*Layering diagram courtesy Kenn Hussey

 Knowledge Representation / Management for Large Scale Applications
 Provide broad metadata, process, service & asset management facilities (including
feedback/lessons learned…)
 Enable rich cross-domain, cross-process, cross organizational modeling supported by
mapping & transformation services to provide maximum flexibility, interoperability
 Leverage standards and best practices in information architecture, metadata modeling,
management, registration, and governance, and asset management & registration
 Provide incremental reasoning capabilities for model validation, transformation services

 Repeatable, reusable, interoperable

+ Hypothetical Conceptual Model “EU-Rent”
20 20

Produced using Embarcadero EA/Studio Business Modeler Edition, courtesy Kenn Hussey


+ Hypothetical Logical Model “EU-Rent”
21 21

Produced using Embarcadero ER/Studio, courtesy Kenn Hussey

+ Hypothetical Physical Model “EU-Rent”
22 22

Produced using Embarcadero ER/Studio, courtesy Kenn Hussey


+ Classifying ontologies
23 23
Classification techniques are as diverse
as conceptual models; and generally
include understanding
KR System
 Level of Expressivity
 Level of Complexity / Structure
OO Software Model
 Granularity
Entity – Relationship
 Target Usage, Relevance Model
 Amount of Automation, Reasoning Requirements Concept Map
 Prescriptive vs. Descriptive / Reliability / Level Topic Map
of Authoritativeness

Level of Expressivity
 Design Methodology Database Schema

 Governance XML Schema

 Vocabulary Management, Metrics Hierarchical Taxonomy
Simple Taxonomy
Glossary

Level of Complexity

+ Framework of dimensions
24 24
 Semantic Dimensions
 Expressiveness: represents how well a KR language addresses
increasingly complex semantics
 Structure: represents how well an ontology encodes semantics, with
the same or less expressivity than the KR language
 Granularity: represents the level of detail specified in an ontology
 Pragmatic Dimensions
 Intended use: the original use case(es), or purpose for developing a
particular ontology
 Automated reasoning: the extent to which the ontology is designed
to be used for automated reasoning
 Prescriptive vs. Descriptive: the extent to which an ontology was
intended to be used for descriptive purposes vs. normative
prescriptive use (i.e., with high degree of concern for correctness)
Reference: http://ontolog.cim3.net/cgi-bin/wiki.pl?OntologySummit2007


+ Model dynamics
25 25
Model centric perspectives characterize the ontologies
themselves and are concerned with their structure, formalism
and dynamics.
Perspective One Extreme Other Extreme
Level of Least authoritative, broader Most authoritative, narrower,
Authoritativeness shallowly defined ontologies more deeply defined ontologies
Source of Passive (Transcendent) - Active (Immanent) - Structure
Structure Structure originates outside emerges from data or behavior
the system
Degree of Informal or primarily Formal, having rigorously
Formality taxonomic defined types, relations, and
theories or axioms
Model Dynamics Read-only, ontologies are Volatile, ontologies are fluid &
static changing
Instance Read-only, resource instances Volatile, resource instances
Dynamics are static change continuously


+ Application characteristics
26 26

Application centric perspectives are concerned with how
applications use and manipulate ontologies.

Perspective One Extreme Other Extreme
Control/Degree Externally focused, public Internally focused, private
of Manageability (little or no control) (full control)
Application Static (with periodic updates) Dynamic
Changeability
Coupling Loosely-coupled Tightly-coupled
Integration Focus Information integration Application integration
Lifecycle Usage Design Time Run Time


+ Language evaluation
27 27
KR Expressivity Requirements Reasoning Requirements
Classes
Exceptions
Slots/Attributes
Automatic Classification
Metaclasses
Number Restrictions

Complex Class Extensions (union, intersection) Inheritance
Subsumption Hierarchies Monotonic, Nonmonotonic
Value Restrictions Simple, Multiple

Behaviors, Procedural Definitions, Methods Procedural Support, Execution
Relations / Functions
Slots/Attributes
Subsumption Hierarchies Constraint Checking, Deduction
Built-in Functions, Equations, Formulae
Instances / Individuals / Facts
Axioms
Reasoning with Rules
Production Rules (forward and backward chaining)

* Derived from “Evaluating Knowledge Representation and Reasoning Capabilities of Ontology Specification Languages”,
Oscar Corcho and Asuncion Gomez-Perez, January 2002


+ Considerations
 Intended use of ontologies, including domain requirements (e.g., 28 28
scientific and engineering apps require formulas, units of measure,
computations that may be challenging to represent)
 Intended use of KRSs that implement them, including reasoning
requirements, questions to be answered
 For distributed environments, the number and kinds of resources,
processes, services requiring ontologies – how distributed, how
unique, developed collaboratively or independently, dynamic
community participation or static
 What kinds of transformations are required among processes,
resources, services to support semantic mediation
 Ontology and KRS alignment / de-confliction / ambiguity resolution
requirements
 Ontology and KRS composition requirements, dynamic vs. static
composition, in what environment and under what constraints
 Performance, sizing, timing requirements of target environment


+ A Little Methodology …
29 29
 Requirements, domain & use case analysis are critical
 Develop initial source/reference material
 Focus on system or application requirements
 Iterative development starting with a “thread” that covers basic
capabilities can ground the work and prioritize decisions
 Need to understand and communicate
 Architectural trade-offs, cost & technical benefits
 The nature of the information & kinds of questions that need to
be answered drive the architecture, approach, and ontology
scoping and design
 Reuse standards and well-tested, available ontologies
whenever possible


+ What to look for
 A controlled vocabulary
30 30
FAA & IATA airport codes, ACRISS car codes, …
 A hierarchical or taxonomic structure (for query expansion)
Vehicle, Ground-based Vehicle, Wheeled Vehicle, Powered Vehicle, Automobile,
Sedan…
 Knowledge supporting structured queries
Find all available hybrid SUVs or hybrid sedans that can seat four adults within a
reasonable taxi ride of Planet Hollywood, Las Vegas for 3-4 days the week of
June 20th
 Efficient inference (i.e., limited expressive power) vs. increased
expressivity (potentially expensive or resource bounded
computation)
 Custom reasoning for temporal relations, geospatial, dynamic
algorithm / equation evaluation, process-specific, conditional
operations
 Computational tractability


+ Start with canonical definitions
 General concepts as well as domain-specific knowledge 31 31
 Basic starting point – cross-domain definitions
 Namespace definitions, metadata, naming conventions, governance policies
 Commonly used structures & vocabularies, such as domain-specific
vocabulary & messaging standards, international country & language codes
(ISO), national postal addressing or other government standards, industry
best practices
 Common metadata for ontology & schema management (e.g., Dublin Core,
for documents & models, ISO 1087 for synonyms & similar relations, MIME
media types for images & multimedia, etc.)
 Domain vocabularies must be prioritized, selected based on business
requirements, clear ROI
 Common early targets include
 Smart search (pull); customer experience & cross-sell / upsell (push)
 Richer interoperability among trading partners
 Service registration, description, discovery & management
 Asset/artifact repository search & retrieval
 Automated verification


+ IDEF5 (Integrated Definition Methods) Ontology
Capture Method Analysis
32 32
 Layout a high-level architecture for ontology elements
 Identify the relationships among elements – roles, domain,
interface, process, utility
 For each ontology element
 Describe its domain and scope, how it will be used
 Identify example questions and anticipated/sample answers for the
application(s) it will support
 Identify key stakeholders, ownership, maintenance, resources for
instance knowledge
 Describe anticipated reuse/evolution path
 Identify critical standards, resources that it must interoperate with,
dependencies
 Resources
 http://www.idef.com/IDEF5/html
 http://www.kbsi.com/technology/methods/sbont.htm


+ Principles & Methods for Terminology Work
(ISO 704)
33 33
 Methodology for describing concepts & terms
 Uses ISO 1087 for terminology
 Uses ISO 860 for terminology “harmonization” (alignment) methods
 Basis for typical methods used for taxonomy development today
 Describes how to flesh out definitions
 Recommendations strategies for relating terms to one another
using standard vocabulary
 ISO 1087 – great resource for language to describe kinds of
relationships, acronyms & other designations, preferred vs.
deprecated terms, etc.
 ISO 860 augments this with recommendations for vocabulary
comparison


+ Part 2: OWL Basics

 A quick intro to RDF 34 34

 Basic OWL constructs: classes, properties & slots
 Inheritance & constraints
 Class & slot inheritance
 Property, slot & slot value constraints
 Individuals & literals
 Disjoint & equivalence relations
 New features of OWL 2
 A few rules of thumb
 Depth vs. breadth, naming, synonymous terms
 Class vs. property value, class vs. individual
 Inverse properties
 Limiting scope

+ Semantic Web
35 35
"The Semantic Web is an extension of the current web in which
information is given well-defined meaning, better enabling computers
and people to work in cooperation."
-- Tim Berners-Lee


+ Guiding Principles
36 36
 Historically, knowledge representation and reasoning systems have
operated under closed world assumptions
 Uncertainty is magnified under open-world, “wild, wild web” conditions,
making reasoning much more difficult
 Semantic web languages are designed to support less certainty, to
provide “better” search results, informed answers to questions, not
absolutes
 Because they are based on XML, such languages can assist businesses in
leveraging existing investment in mark-up, content, and data
 To augment business intelligence/analysis and knowledge mining
 To support knowledge sharing and collaboration, augment enterprise
information integration
 Enrich web services and other applications
 Support policy-based applications and ensure compliance with policy

at a lower cost with higher potential ROI than traditional computing
methods


+ Resource Description Framework (RDF)
37 37

 Describes relationships
 Uses URIs used for naming
 Language has
 graph based model
 RDF/XML serialization (exchange syntax)
 other presentation syntaxes (N3, Turtle, …)

 Specification, W3C presentations, tools are available at
 Semantic Web: http://www.w3.org/standards/semanticweb/
 Linked Data: http://www.w3.org/standards/semanticweb/data
 RDF: http://www.w3.org/standards/techs/rdf#w3c_all

 New RDF “next steps” revision working group under way to
make specific enhancements, “fixes” to the standard
 RDF Working Group: http://www.w3.org/2011/rdf-wg/wiki/Main_Page


+ RDF Notation Options
38 38
Subject

http://semtech2011.semanticweb.com/travel.owl#Conference

Predicate
Graph: http://semtech2011.semanticweb.com/travel.owl#heldAt

http://semtech2011.semanticweb.com/travel.owl#HiltonSFUnionSquare
Object

<rdf:Description rdf:ID=“Conference"/>
XML/RDF: <semtech:heldAt rdf:resource="#HiltonSFUnionSquare"/>
</rdf:Description>

N3: semtech:Conference semtech:heldAt semtech:HiltonSFUnionSquare .


+ RDF Schema (RDFS)
 An RDF vocabulary that provides for identifying: 39 39

 classes,
 subsumption (inheritance) relations for classes,
 subsumption (inheritance) relations for properties,
 domain and range for properties


+ RDF Schema (RDFS)
40 40
semtech:Hotel
rdfs:subClassOf
Graph: rdf:type

rdfs:Class semtech:ConferenceHotel
rdf:type
rdf:type

semtech:HiltonSFUnionSquare

<rdf:Description rdf:ID="Hotel">
XML/RDF: <rdf:type rdf:resource="http://www.w3.org/2000/01/rdf-schema#Class"/>
</rdf:Description>

<rdfs:Class rdf:ID=“ConferenceHotel">
<rdfs:subClassOf rdf:resource="#Hotel"/>
</rdfs:Class>

<semtech:ConferenceHotel rdf:ID=“HiltonSFUnionSquare“/>


+ The Web Ontology Language (OWL)
 Two languages emerged in parallel to address semantic web 41 41
requirements
 DAML-ONT, supported by the DARPA/DAML program
 OIL (Ontology Inference Layer) developed by EU & US researchers
 Merged DAML+OIL was submitted to the W3C 2002, formed the
basis for the WebOnt Working Group
 OWL extends RDF Schema
 Has an RDFS based syntax and reuses some RDF vocabulary (e.g.,
subClassOf, domain, range)
 Adds rich primitives and redefines others (transitivity, inverse, cardinality
constraints, complex class definitions)
 Describes the structure of a domain in terms of classes and
properties
 Uses RDFS for class/property membership assertions (ground
facts), XML Schema Datatypes
 OWL specifications became W3C recommendations in February
2004
 OWL 2 specifications was adopted in October 2009

+ General nature of descriptions
42 42
a WINE

a LIQUID General Categories
a POTABLE

grape: chardonnay, ... [>= 1]
sugar-content: dry, sweet, off-dry Structured
color: red, white, rose Components
price: a PRICE
winery: a WINERY

grape dictates color (modulo skin) Interconnections
harvest time and sugar are related Between Parts


+ General nature of descriptions
43 43
Class a WINE

Superclass a LIQUID General Categories
a POTABLE
Number / Card
Restrictions grape: chardonnay, ... [>= 1]
sugar-content: dry, sweet, off-dry Structured
Roles / color: red, white, rose Components
Properties price: a PRICE
winery: a WINERY
Value
Restrictions grape dictates color (modulo skin) Interconnections
harvest time and sugar are related Between Parts


+ Description development
44 44

 Define domain terms and inter-relationships
 Define concepts in the domain (classes, nouns)
 Identify subclass/superclass relationships
 Identify attributes/properties/slots (verbs)
 Identify any general properties (relations, functions, verbs)
 Restrict slot values

 Define a representative set of individuals – for testing & evaluation,
prototype development
 Define individuals (instances)
 Define relationships between individuals (filling in slots)


+ Classes & class hierarchy
45 45
 A class is a concept in the domain
 Vintage – a wine made from grapes grown in a specified year
 A class of properties (flavor, body, color, sugar…)
 A class is a collection of elements with similar properties
 White wine – wines made from white grapes
 White table wine – wines made from white grapes that are not
appellations or regional (not “quality wine” in the EU)
 A class contains necessary conditions for membership (specific
varietal, field, micro-climate, date picked, date bottled)
 Instances of classes
 Andrew Murray 2005 Late Harvest Viognier
 Robert M. Parker, Jr,’s Wine Advocate review, dated February 28, 2002
 Los Olivos, Santa Barbara County, California, USA


+ Class inheritance
46 46
 Classes are organized into subclass-superclass (or
generalization-specialization) hierarchies
 True subclass relationships are the basis of a formal is-a
hierarchy
Classes are “is-a” related if an instance of the subclass is an
instance of the superclass
 Classes may be viewed as sets
 Subclasses of a class are comprised of a subset of the
superset
 Examples
 RedWine is a subclass of Wine
Every red wine is a wine or every instance of a red wine (e.g.,
Marietta Old Vines Red) is an instance of wine
 NapaValleyWine is a subclass of CaliforniaWine
Every wine from Napa Valley is a wine from California


+ Levels in the class hierarchy
47 47
 Different modes of development
 Top-down - define the most general concepts first and then
specialize them
 Bottom-up - define the most specific concepts and then organize
them into more general classes
 Combination (typical – breadth at the top level and depth along a
few branches to test design)
 Class inheritance is transitive
 A is a subclass of B (white wine, dessert wine are subclasses of
wine)
 B is a subclass of C (viognier is a subclass of white wine, late
harvest wine is a subclass of dessert wine)
 therefore A is a subclass of C (late harvest viognier is a subclass of
white wine, dessert wine, & wine)


+ Classes in OWL 2
48 48

 6 kinds of class expressions in OWL
 5 anonymous
 Definitions can be nested using set-theoretic constructs

+ Properties & slots
49 49
 Slots in a class definition describe attributes of members of a
class
Each wine will have color, sugar content, flavor, body, etc.
 Types of properties
 “intrinsic” properties: flavor and color of wine
 “extrinsic” properties: name and price of wine
 parts: ingredients in a recipe
 relations to other objects: producer of wine (winery)
 Data and object properties
 simple (datatype) contain primitive values (strings, numbers)
 complex properties may contain other objects (e.g., a winery
instance)


+ Properties in OWL 2
50 50


+ Class & slot inheritance
51 51
 A subclass inherits all the slots from its super class
If a wine has a name and flavor, a red wine also has a name and flavor
 If a class has multiple super classes, it inherits slots and
restrictions from all of them
Latin is both an individual language and an ancient language. It inherits “has
attested literature: true” from ancient language and “has indigenous
name: lingua latina” from individual language


+ Property (slot) constraints / restrictions
 Number restrictions describe or limit the number of possible values
a particular property can have 52 52
 A language must be associated with at least one English name and at least
one French name
 A language may be associated with zero or more Indigenous names
 Cardinality – similar meaning to classical set theory, measures the
number of elements in the set (restriction class)
 Cardinality – cardinality N means the class defined by the property
restriction must have exactly N values (individual or literal values)
 Minimum cardinality - 1 means that there must be at least one value
(required), 0 means that the value is optional
 Maximum cardinality - 1 means that there can be at most one value
(single-valued), N means that there can be up to N values (N > 1, multi-
valued)


+ Slot value constraints
53 53

 Slot value type – defines the set of possible values for the
property
 String: a string of characters (“McGinley Vineyard”)
 Number: an integer or a float (3.5 acres)
 Boolean: a true/false flag
 Enumerated type: a list of allowed values (red, white, rose)
 Filler: a single value (e.g., the color slot for a RedWine must be
filled with the single value “red”)
 Object type – a class defined in an ontology (e.g., Winery is the
value restriction on the hasMaker slot on the class Wine)


+ Domain & range properties
54 54
 In OWL and many other KR languages, relations (properties, slots) are
strictly binary
 The domain & range represent the source & target arguments,
respectively, for the property
 Domain of a slot – the class (or classes) that may have the slot − Wine
is the domain of the slot hasWineColor
 Range of a slot – the class (or classes) to which slot values belong −
everything that fills the hasWineColor slot is an instance of the
enumerated class {red, white, rose}
 Some KR languages that inherently support n-ary relations, such as
CL, do not make this distinction
 More flexible, intuitively more like mathematics, where functions have
ranges (or return types) but not all relations are functions
 Requires additional relations to specify argument order, which can be
critical for ontology alignment


+ Property inheritance
 A subclass inherits all the slots of its superclass(es) 55 55

 A subclass can add constraints to “narrow” the set of allowed
values
 Make the cardinality range smaller
 Replace a class in the range with a subclass


+ Individuals
56 56
 An Individual (instance, object in other paradigms)
 Any class that an individual is a member of, or is an individual of, is a type
of the individual
 Any superclass of a class is an ancestor (or type) of the individual
 Specify slot values for the individual
 Slot values should conform to the constraints such as range, value type,
cardinality restrictions, etc.


+ OWL Individuals
57 57
Graco Garmin

Chevrolet
Sony
BMW


+ OWL Statements
58 58
Graco Garmin

Chevrolet
Sony
BMW

a Mini
Cooper
a Camaro LT
Convertible


+ OWL ObjectProperty
59 59
Graco Garmin

Chevrolet
Sony
<owl:ObjectProperty rdf:ID="builtBy">
BMW
<rdfs:range rdf:resource="#Enterprise"/>
<rdfs:domain rdf:resource="#DurableGood"/>
<owl:inverseOf rdf:resource="#hasBuilt"/>
</owl:ObjectProperty>

a Mini
Cooper
a Camaro LT
Convertible


60 60
range Graco Garmin

Chevrolet
Sony
BMW
<owl:ObjectProperty rdf:ID="builtBy">

a Mini
Cooper
a Camaro LT
Convertible


61 61
range
Graco Garmin

Chevrolet
Sony
<owl:ObjectProperty rdf:ID="builtBy"> BMW

a Mini
domain Cooper
a Camaro LT
Convertible


+ Inverse Properties
62 62
domain Graco Garmin

Sony Chevrolet

<owl:ObjectProperty rdf:ID="hasBuilt"> BMW
<rdfs:range rdf:resource="#DurableGood"/>
<rdfs:domain rdf:resource="#Enterprise"/>
<owl:inverseOf rdf:resource="#builtBy"/>

a Mini
range Cooper
a Camaro LT
Convertible


+ OWL Class
63 63
Graco Garmin

Chevrolet
<owl:Class
Sony
rdf:ID="ManufacturingEnterprise"/>
BMW

<owl:Class rdf:ID="DiscreteManufacturingEnterprise">
<rdfs:subClassOf>
<owl:Class rdf:about="#ManufacturingEnterprise"/>
</rdfs:subClassOf>
</owl:Class>


+ Class Axioms
64 64

A ⊆ B where B
 Subsumption (necessary)
B


is a class description
partial or primitive class A

 Definition (necessary and sufficient)
 C ≡ D where D
D
is a class description
complete or defined class
C

Courtesy Evan Wallace, NIST


+
Meta-Properties – Global Cardinality Restriction
65 65
Range
 Functional

Domain
Range

 Inverse Functional

Domain

+ Class Descriptions – Property Restriction
66 66

property P

Individual
of Class C

 Quantified property restriction (type)
 Universally quantified – allValuesFrom
 Existentially quantified - someValuesFrom

 hasValue property restriction (value)

 Property cardinality restriction (# of values)


+ Disjoint classes
A 67 67

B
 Classes are disjoint if they cannot have common instances
 Disjoint classes cannot have any common subclasses
 If winery and wine are disjoint, then there is no instance that is
both a winery and a wine; there is no class that is both a
subclass of winery and a subclass of wine
 Disjointness is often used to aid consistency checking
 Disjointness is also helpful in teasing out subtle distinctions
among classes across multiple ontologies
 Equivalence is also often used to identify the same concepts
across ontologies that may be named differently, or to name
classes defined through class axioms


+ New features in OWL 2
68 68
 New property constructs
 Reflexive, irreflexive, & asymmetric properties
 Property chains
 Disjoint properties
 Keys
 Self-reflexive properties
 Qualified cardinality restrictions
 Negative property assertions

 New class axioms
 Disjoint unions
 Disjoint classes


+ More new features
 Extended datatype handling
69 69
 Better support for XML Schema Datatypes
 New datatypes for OWL specifically (owl:real, owl:rational, rdf:PlainLiteral)
 Custom datatype definition & use of xsd facets / datatype restrictions

 New data range restrictions
 Intersections, unions, complements

 Support for punning
 Limited support for a particular element to be defined as both a class &
individual, for example

 Improved support for annotations (e.g., metadata about ontologies,
provenance, transformation detail, etc.)

 See http://www.w3.org/TR/owl2-new-features/ &
http://www.w3.org/TR/owl2-quick-reference/ for detail

 Current work specifically focused on provenance:
http://www.w3.org/2011/prov/wiki/Main_Page

+ Rules of Thumb
70 70

 Cycles are common in many KR
systems, though rarely “a good thing”
 Cycles are disallowed by some tools
because they prohibit “code
generation”, export -- including
RDF/OWL
 Classes A, B, and C have equivalent
sets of instances
 By many definitions, A, B, and C are
equivalent
 Use owl:equivalentClass instead of
creating cycles


+ Siblings in the class hierarchy
71 71

 All siblings should be specified at
roughly the same level of generality
 Compare to section and subsections in
a book


+ Class specification
72 72
 If a class has only one child, there
may be a modeling problem – often a
sign that a definition is incomplete
 If the only Red Burgundy we have is
Côtes d’Or, why introduce the
subclass?

 Subclasses of a class usually have
 Additional properties
 Additional slot restrictions
 Participate in different relationships
 Compare to bullets in a bulleted list


+Creating levels & subclasses
73 73

 If a class has a large number of
subclasses, it may be useful to
define intermediate levels
 For example, in the domain of
wines, there are natural
groupings around wine color
 However, if no natural
classification exists, the long list
may be appropriate


+ Inheritance, naming, synonyms
74 74
 A “wine” is not a subclass of “wines”

 A particular vintage should be classified as
an instance of the class Wines
Class
 Class names should be either
 all singular
 all plural
instance-of
 Synonym names for the same concept are not
Instance different classes
MariettaOldVinesRed
 Many systems, metadata standards support
synonymous terms as part of a class definition
 OWL allows defining necessary and
sufficiency condition definitions thereby
allowing synonym definitions to be “first
class” terms

+ Class vs. property value
75 75

 Do concepts with different slot values become restrictions for
different slots?
 How important is the distinction for the domain?
 Class definitions for most domains should be fairly stable – i.e.,
they should not change frequently once the definitions are
established and individuals created


+ Class vs. individual
76 76

 Individual instances are the most specific objects in an ontology
 If concepts form a natural hierarchy, represent them as classes
 If they will have instances below them, represent them as classes


+ Limiting scope
77 77

 An ontology should not contain all the possible information
about the domain
 No need to specialize or generalize more than the application
requires
 No need to include all possible properties of a class
• Only the most salient properties
• Only the properties that the application requires

 Ontologies of wine, food, and their pairings probably will not
include details such as:
 Bottle size (half bottle, full bottle, magnum, …)
 Label color
 Wine bottle color (green, amber, …)


+ Part 3: Putting It All Together
78 78
 Syntax Checking
 Consistency Checking
 Instance Data Analysis & Evaluation Services
 Applications
 Lighter Weight
 Richer Applications
 Examples – showing a combination of semantic web
components in action: TW Wine Agent
 Understanding, Trust & Proof: InferenceWeb
 eScience Applications: Virtual Solar Terrestrial Observatory,
SONet, & PopSciGrid
 Leveraging Explanations in a Cognitive Assistant
 Questions

+
Syntax checking
 For RDF & OWL Ontologies 79 79
 RDF syntax checking, graph visualization
• W3C RDF Validator (http://www.w3.org/RDF/Validator/)
• Jena API & Toolkit (http://jena.sourceforge.net/)
 OWL syntax checking, OWL dialect
determination
• OWL Consistency Checker
(http://clarkparsia.com/pellet/, also provides full OWL 2
DL reasoning capabilities)
• OWL 2 Validator (Univ. of Manchester,
http://owl.cs.manchester.ac.uk/validator/ )
• Protégé OWL (http://protege.stanford.edu/)
• Jena API & Toolkit (http://jena.sourceforge.net/)

 Every tool provides unique capabilities;
sophisticated projects may require multiple
approaches
 Tools listed are open source, commercial
options are also available

+ Consistency checking
 Requirements are typically application specific 80 80
 RDF vocabularies should be checked by an RDF validator or rule engine
 Jena Semantic Web Framework - http://jena.sourceforge.net/
 Pychinko, MindSwap’s Rete-based RDF-friendly rule engine – (CWM clone)
http://www.mindswap.org/~katz/pychinko/
 OWL Ontologies should be run through a consistency checking reasoner
 Pellet (open source, originally from Mindswap, supported by Clark & Parsia, LLC)
– http://clarkparsia.com/pellet/,
 RacerPro – http://www.racer-systems.com/
 FaCT++ – http://owl.man.ac.uk/factplusplus/
 HermiT OWL Reasoner – http://www.hermit-reasoner.com/
 KAON2 infrastructure for managing OWL-DL, SWRL, and F-Logic ontologies –
http://kaon2.semanticweb.org/
 VIStology's ConsVISor OWL Consistency checker –
http://68.162.250.6:8080/consvisor/

 OWL instance data may also be run through checking tools
 TW Instance data evaluation - http://onto.rpi.edu/demo/oie/


+ Example process
81 81

Pellet OWL DL Reasoner

Valid OWL
(DL consistency checked, concept
satisfiability checked)

TW Instance Data
Evaluation

Valid OWL
(FOL consistency checked,
deductive closure)

Valid RDF/XML Documents
Ontology Registry (Valid OWL Syntax Representation)
& Library Application Environment

Ontologies should be checked to ensure consistency,
limit potential for invalid conclusions


+ Inconsistencies caused by distributed OWL data
82 82

Wine ontology
wine:Wine
rdfs:subClassOf rdfs:subClassOf

wine:EarlyHarvest wine:LateHarvest
owl:disjointWith

A new ontology A new instance data
rdfs:subClassOf rdfs:subClassOf rdf:type rdf:type

wine:BadWineClass wi:BadWineInstance

Case1: semantic inconsistency Case2: semantic inconsistency
caused by a new class. caused by a new instance.

+ TW OWL instance-data evaluation
83 83
 Motivation
 Objective metrics are needed to evaluate if semantic web instance
data is ready for practical use
 Next generation Chimaera for more diverse and more instance-
oriented applications

 Challenges
 Criteria for “ready for use” vary across applications
 Existing syntax and logical-consistency checking is not enough for
applications using some standard practical assumptions (e.g.
closed world, unique names, …)

 Solution
 Extensible & customizable service-oriented architecture
 Integrated environment for evaluating issues beyond standard
syntactic correctness and logical consistency

 See http://onto.rpi.edu/demo/oie/


+ Some issues in OWL instance data
84 84

owl:cardinality owl:cardinality "1"
"1"
owl:onProperty owl:onProperty
hasMaker hasColor
rdf:type rdf:type
owl:Restriction owl:Restriction
rdfs:subClassOf rdfs:subClassOf

Wine Wine hasColor Red
W
W rdf:type
MPV rdf:type EPV hasColor White

 Missing property value (MPV)
 Excessive property value (EPV)


+ “Pluggable” instance evaluation services
<?xml version="1.0"?>
<!DOCTYPE rdf:RDF [
<!ENTITY rdf "http://www.w3.org/1999/02/22-rdf-syntax-ns#" >
<!ENTITY wine "http://tw.rpi.edu/2008/04/wine-lite.owl#" > ]>
85 85
<rdf:RDF xmlns:wine = "&wine;" xmlns:rdf = "&rdf;">


<wine:Zinfandel rdf:about="#W1">
<wine:hasColor rdf:resource="&wine;Red"/>
<wine:hasFlavor rdf:resource="&wine;Moderate"/>

<wine:hasSugar rdf:resource="&wine;Dry"/>
<wine:hasBody rdf:resource="&wine;Full"/>
</wine:Zinfandel>


<wine:Zinfandel rdf:about="#W3">
<wine:hasFlavor rdf:resource="&wine;Moderate"/>
<wine:hasSugar rdf:resource="&wine;Dry"/>
<wine:hasBody rdf:resource="&wine;Full"/>
<wine:hasMaker>

<wine:Winery rdf:about="#Elyse" />
</wine:hasMaker>
</wine:Zinfandel>

</rdf:RDF>

+ Ontology-enhanced applications
 Simple ‘ontologies’, taxonomy driven applications 86 86
 Controlled, shared vocabularies (content management, faceted search)
 Web site organization & navigation, expectation setting
 Browsing & search engine optimization via mapping to tagged structures such as
the new schema.org from Yahoo!, Google & Bing, Good Relations ontology – see
http://purl.org/goodrelations/
 “Smarter” search support (query expansion such as FindUR, e-Cyc; SPARQL for
linked data applications)

 More expressive ontologies provide more options for
 Configuration, e.g., AT&T PROSE/Questar
 Data Integration, e.g., Virtual Observatories (VSTO, …)


Semantic Sommelier
+
87 87

Semantically-enabled
advisors, utilizes

 ontologies

 reasoning

 social

 mobile

 provenance

 context


+ TW Wine Agent – semantic interoperability
Wine Agent receives a meal description and retrieves a 88 88
selection of matching wines available on the Web, using
an ensemble of standards and tools
 RDF & OWL − for encoding wine/food listings and pairing
recommendations
 Semantic MediaWiki − for publishing user-contributed
recommendations*
 Twitter and Facebook − for posting recommendations
 Pellet − for deriving knowledge from wine ontology and
recommendations
 SPARQL − for querying wine/food listings with
recommendations
 Inference Web − for explaining TW Wine Agent’s intelligent
behavior

+ Social Semantic Web data publishing
89 89

Collaborative wine recommendations
 Using Semantic MediaWiki to
manage users and user-contributed
food/wine pairings
recommendations
 Using semantic form to add OWL
Copyright © 2011 Thematix & McGuinness Associates instance data

+ Semantic Query:
food hierarchy & recommendation
90 90


+ Wine Agent processing
91 91

Given a description of a meal
 Combine wine ontology and the
OWL data published at Wiki
 Use Pellet and SPARQL to state a
premise (the meal) and query
the knowledge base for a
suggestion for a wine
description or a set of wine
instances
 Use Inference Web to explain
results (descriptions, instances,
provenance, reasoning engines,
etc.)


+ Wine Agent for iPhone
92 92

 Client application which talks to a SW
service

 Make requests for dishes and wines
using auto-generated interfaces

 Make recommendations to the system
for others


+ Dynamic interfaces as function of ontology
93 93

owl:Functional
or
rdfs:range == owl:maxCardinality == 1
owl:Class

rdfs:range != owl:Class U
rdfs:Literal

rdfs:range ==
rdfs:Literal Otherwise


+ Descriptions
94 94
 Resulting descriptions are
converted to classes (for creating
recommendations) or to instances
(for obtaining recommendations)

 When going to a class, all “Is A”
statements become
rdfs:subClassOf and all other
properties become owl:Restrictions

 Instances are realized by the
reasoner to identify appropriate
recommendations


+ Getting the Recommendation
95 95
 Recommendations are made up of
two classes: 1 dish, 1 wine

 When the instance is classified,
the agent returns the matching
recommendations

 Tapping a particular
recommendation causes the wine
agent to look for items belonging
to the matching pair


+ Extensibility
96 96
 Wine Agent Ontology is extensible
 “Menu file” in RDF

 Restaurants can introduce new
classes and instances to provide
better matches for users

 Searches can be restricted to
imported menus


+ Future Work
97 97

 Group recommendations using multiple iPhones/iPod
touches

 Recommendations based on personal dietary needs and
preferences

 ‘iPellet’, an OWL Reasoner designed for iPhone, for
offline reasoning


+ Observations from the Wine Agent
98 98
 Background knowledge is reasonably simple and built in OWL
(includes foods and wine and pairing information similar to the OWL
Guide, Ontology Engineering 101, CLASSIC Tutorial, …)
 Background knowledge can be used for simple query expansion over
wine sources to retrieve for example documents about red wines
(including zinfandel, syrah, …)
 Background knowledge used to interact with structured queries such
as those possible on wine.com
 Constraints allows a reasoner like Pellet to infer consequences of the
premises and query.
 Explanation system (Inference Web) can provide provenance
information such as information on the knowledge source
(McGuinness’ wine ontology) and data sources (such as wine.com)
 Services work could allow automatic “matchmaking” instead of hand
coded linkages with web resources

+ Trust & Understanding
99 99
If users (humans and agents) are to use, reuse, and integrate system
answers, they must trust them.
System transparency supports understanding and trust.
Even simple “lookup” systems benefit from providing information
about their sources.
Systems that manipulate information (with sound deduction or
potentially unsound heuristics) benefit from providing
information about their manipulations.

Goal: Provide interoperable infrastructure that supports explanations of
sources, assumptions, and answers as an enabler for trust.


+ Explanations, Proof Analysis
100
100

 Framework for explaining question answering tasks
 Stores and manages meta-information about proofs and
explanations through a distributed repository
 Uses the Proof Markup Language (PML) for proof interchange
(OWL-based)
 Services include proof presentation, strategy-based views,
filtering, trust, combination, expansion, checking, search, and
other capabilities
 Integrated browsing and display of PML documents from
diverse sources
 Rewriting capabilities for improved understanding
 Multi-modal dialogue options including alternative strategies
for presenting explanations, visualizations, and summaries


+ Inference Web (IW)
101
101

End
Users

End-User Data Access &
Validate published PML data
Interaction Data Analysis
services Services
Explanation via Graph

Distributed
Explanation via Summary PML data
Explanation via Annotation Access published PML data

Inference Web is a semantic web-based knowledge provenance management
infrastructure:

• PML for encoding and interchange provenance metadata in distributed environment
• Interactive explanation services for end-users
• Data access and analysis services for enriching the value of knowledge provenance

+ Proof Markup Language (PML)
 A new kind of linked data on the Web 102
102

World Wide Web D
PML
PML data
Enterprise Web data
PML PML PML
data D PML
data data data

D Enterprise Web
D D D D PML
D data
…

 Modularized & extensible
 Provenance: annotate provenance properties
 Justification: encodes provenance relations
 Trust: add trust annotation

 Semantic Web based

+ Technical Architecture
IW Registrar
call Web Service Edit edit 103
103
PML database
Interface Interface

machine agents generate domain experts
generate
edit
PML API
translate
Proofs PML Translator
(N3, KIF)
PML (provenance, justification, trust)
Computation Tools
(validation, conflict detection, abstraction)
is-stored-as
Data Access
Presentation Tools
(IW Browser) PML Java PML parse PML Web
Objects API Documents
reads uses harvests & indexes
searches harvests
IW Search subscribes


Making Systems Actionable
using Knowledge Provenance
Mobile CALO
Wine Agent
104
104

Combining
Proofs in TPTP

Intelligence Analyst Tools

Knowledge
Provenance
in Virtual
GILA NOW including Data-gov Observatories
Copyright © 2011 Thematix & McGuinness Associates 104

+ Example Proof Tree Browser
105
105


+ Explaining Extracted Entities
Source: fbi_01.txt (in NL) 106
106
Same conclusion Source Usage: span from 01 to
from multiple 78
extractors

Conflicting
conclusion from one
extractor (with
annotations)

This extractor decided
that Person_fbi-01.txt_46
is a Person and not
Occupation (in logical
format – e.g. KIF)


+ Trustworthiness of Ramazi report
From FBI
107
107

From intercepts

From CIA


Ontology 101 - Kendall & McGuiness

Ontology 101 - Kendall & McGuiness

Recommended

Recommended

More Related Content

Similar to Ontology 101 - Kendall & McGuiness

Similar to Ontology 101 - Kendall & McGuiness (20)

Recently uploaded

Recently uploaded (20)

Ontology 101 - Kendall & McGuiness