Knowledge Management Chapter 9 Ontologies

UoCalgary 2003

c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Recommended References • D.B. Lenat, R.V. Guha: Building Large Knowledge Based Systems. Addison-Wsesley 1990. • B. Bachmann: A solution for the semantic unification problem to reuse knowledge-based systems. Infix Verlag, Reihe DISKI Nr. 162, 1997 • R.J. Brachmann: The future of knowledge representation. Proc. AAAI 1990, S. 1082-1092. • T.Gruber: A translation approach to portable ontology specifications. Knowledge Acquisition 43(1995), S. 907-928

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Ontologies and Knowledge Management • In order to provide knowledge support for agents one first has to collect and to restructure knowledge. • The restructuring has to be done in such a way that al agents participating in the process (or all members of a company) can be supported in an efficient way. • The knowledge comes usually from different sources that have different representation formats as well as a different semantics for the used terms. • The idea of an ontology is to collect, combine and restructure such knowledge.

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

General • Common Language: „People can‘t share knowledge if they don‘t speak a common language.“ T. Davenport

A common language includes: – a common set of symbols (e.g. words to be used) – a common set of concepts (terms), with a defined meaning – a common set relationships among the concepts, with a defined meaning

• The very same language may be used by different people: – As writers – As readers

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Characteristics “An ontology is a formal, explicit specification of a shared conceptualization”

• An ontology is: – formal: is machine understandable (therefore we need formal languages!) – explicit specification: contains explicitly defined concepts, properties, relations, functions, constraints, axioms, ... – shared: represents consensual knowledge of a community – conceptualization: abstract model of some phenomenon in the world

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

KBSs and Ontologies •

Knowledge based systems are supposed to solve problems. This can be interpreted differently: – Direct use: The system is ready to obtain a problem as in put – Indirect use: The system itself does not solve specific problems but provides a basis to built problem solving systems fast.

•

• • •

Basic idea for the latter: An ontology represents knowledge about a domain independent from the intended use. This can e.g. be the knowledge of some company which builts cars or airplanes. An important part of an ontology is a terminology which provides the basic concepts and the way they have to be used. Besides the terminology an ontology also contains facts about the domain. Such ontologies are often called generic.

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Types of Ontologies • Ontologies can talk about different issues: top-level ontology

domain ontology

task ontology

application ontology

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Tasks Ontologies (1) •

•

• •

More special are task-Ontologies which contain concepts and facts of interest for some application type: – diagnosis-ontologies, planning-ontologies, configuration ontologies... Task ontologies still do not contain special heuristics etc. but the knowledge is selected in a task specific way. Again, also a task ontology is not intended to solve a specific problem directly. Syntax of an ontology: Like in predicate logic (or some fragment) Additions are not a principle extension: – Types (possible structured) – Identifiers – etc.

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Task Ontologies (2) • Example domain: Technical devices; the question is which knowledge of possible ontologies is useful ? • Tasks: •

•

•

Fault diagnosis: – failure probabilities, value of components (is a detailed investigation still meaningful?), structure of components (influences costs of tests),... Design and Planning: – Situation knowledge, action steps, preconditions, costs and resources, consequences, constraints for parameters,... Electronic commerce: – functional parameters (e.g. maximum speed of a car), energy consumed, guarantee on parts, price of components, additional parts available,...

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Different Ontologies for a Common Goal • •

•

• •

Knowledge can be distributed over different ontologies : Different experts see and describe the world from different views; they have their own knowledge stored in different ontologies: – Financial experts, environment experts, regional experts (in particular in global distribution), technical experts (again many possibilities) ..... Each expert talks and thinks about the same real world objects – in different terminologies – in different degrees of abstraction – in different form of details In order to make use of such distributed knowledge a principal problem of communication arises (see chapter on Predicate Logic). Some part of the problem can be solved by using normed terminologies

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Multiple Use of Ontologies • The idea of using one ontology for different purposes has several consequences: – How to interpret the concepts of an ontology in different situations ? – How to translate the concepts of an ontology into different representations of an application ? – How to select the knowledge for some application ?

• The last questions require domain knowledge. The first two questions deal with formalisms and their syntactic properties. This is discussed in the chapter on predicate logic.

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Ontologies: Related Approaches • Conceptualizations of a domain can be achieved in different ways: – – – –

Glossary Thesauri Database Schema Object-Oriented Model (UML)

• These concepts can support ontologies

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Database Schema & OO Models • Conceptual database schema (e.g. ER diagrams) and OO models (e.g. UML analysis diagrams) allow the formal specification of a conceptualization. • ER and UML differ in the modeling primitives • Relation to Ontology definition: – – – –

formal? : YES can be encoded in machine readable form explicit specification: YES: modeling primitives have a clear semantics shared? : Not per se conceptualization? : YES abstract model containing important concepts.

• If shared, they can become an ontology!

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

•

A Glossary is:

Glossary

– A collection of terms (e.g. given at the end of a book) – Each term has a textual definition – The definition may contain links to other terms

•

Example Glossary for UML (OMG-Unified Modeling Language, v1.4) aggregation: A special form of association that specifies a whole-part relationship between the aggregate (whole) and a component (part). See: composition. association: The semantic relationship between two or more classifiers that specifies connections among their instances. analysis: The part of the software development process whose primary purpose is to formulate a model of the problem domain. Analysis focuses on what to do, design focuses on how to do it. Contrast: design.

•

Relation to Ontology definition: – – – –

formal? : NO, definitions not machine understandable explicit specification? : NO not formal, links can be “semi-formal” shared? : YES, goal is to share it among the readers/users of the book. conceptualization? : YES abstract model containing important concepts. - UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Thesauri •

A Thesaurus is: – – – –

•

A collection of words with semantic relations among them different meanings of words are mentioned classification w.r.t. syntactic categories sometimes with explanations of meanings of words

Example: Wordnet http://www.cogsci.princeton.edu/~wn/ – 118.000 word forms, 90.000 word meanings, – syntactic categories: noun, verb, adjective, adverb, preposition, ... – semantic relations: hyperonym/hyponym (=is-a), meronym/holonym (haspart(part-of), synonym, antonym, ...

•

Relation to Ontology definition: – formal? : Partially, meaning not machine understandable – explicit specification: PARTIALLY formal relations – shared? : YES; at least this is claimed. – conceptualization? : YES abstract model containing important concepts.

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Language Problems and Terminologies (1) Language problems in a language (intra-language problems): ● Different level of abstraction ● Synonyms ● Quasi synonyms ● Vague and unprecise terms Task: ● Building and collecting obligatory terminologies. They may have commercial value. ● This is a major knowledge management task !

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Language Problems and Terminologies (2) •

•

• •

Misunderstandings occur in particular if everyday concepts are used in a specific context (in technology, administration or science) Each sufficiently complex company needs a well understood terminology: Otherwise communication is often impossible Globalization enforces the relevance of terminologies If knowledge based systems are involved this becomes even more serious because machines have no common sense which sometimes helps to avoid misunderstandings

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Example: Different Interpretation of Legal Terms Car seat specification

Investing money

Germany Canada USA Mexico Germany Bahamas Hongkong USA

Different safety regulation

Different tax regulations

Requires always a flexible interpretation

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Terminology •

•

The task of a terminology is to give a base for communication about concepts. Concepts may have a formal definition but it is often missing. The lack of a formal definition should not prevent understanding of partners. Terminologies deal often with imprecise concepts. There are three major elements to support communication – Glossaries – Thesauri – Term records are introduced; they are a special kind of frames.

•

• •

The entries can be – Formal or informal (i.e. they are semi-formal representations) – Mandatory or optional. Even formal concepts like those of mathematics need more entries than just the definition, e.g. information in which context they are used. All three elements are primarily but not always intended for the human user. - UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

A Typical Term Rekord (1) • Mandatory part: – – – – – –

Identifier Source and author: May change over time Definition: In an ideal sense mathematically exact Domain: In order to identify the meaning Explanation and context: For a meaningful use Synonyms and quasi synonyms: Terms with the same or almost the same meaning – Abbreviation, short form – Language (German, English,...)

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

A Typical Term Rekord (2) • Optional part: – – – – – –

Phrase record: Typical phrases in which the term is used Key words: For retrieval purposes Images, graphics, diagrams Alternative spelling Translations additional information

• The intention of this part is to improve the understanding and the meaningful use of the term. It depends very much on the term and the specific situation.

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Example (1) • Winding of composite material in mechanical engineering: • From a text book: – Coiling is a procedure which is to a large degree automated for producing hollow bodies of fiber enriched synthetic material. The enriching fibers are coiled around a rotating nucleus which remains in the component or is removed from it, depending on the geometry.

• First step: – Generation of an informal term record.

• Next steps: – Stepwise translation to intermediate term records until the ontorecord is reached.

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Example • Term record: – – – – – – – – – – –

Mandatory part: Term: “Winding” Source: IVW-Technologien.txt Author: G.Schmidt Subject field: Mechanical engineering Context: Manufacturing of hollow bodies Synonyms: Roll, coil Short forms: Unknown Optional part: linguistic Information: Word stem “wind” (verb) notes: Picture X

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Formal Aspects of Ontologies • •

•

• •

Example: identifier is the name of some unary relation (a class), or of an n-ary relation or of a function.Possible axioms: ∀x1,...,xn identifier(x1,...,xn) → Φ (x1,...,xn) ∀x1,...,xn identifier(x1,...,xn) ↔ Φ (x1,...,xn) The first form describes a necessary and the second form a necessary and sufficient condition. Φ is an expression in which identifier does not occur. (The introduction of identifiers corresponds to the axioms in terminological logic; in each ontology a terminology is introduced, see chapter on Predicate Logic). In addition often complex knowledge is represented. Again, it is not intended to represent problem specific strategies for obtaining solutions. The idea is rather to add these later for specific problems and for different purposes.

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Formalizing an Ontology • •

• •

Ontologies need a formal representation in order to be processed by a computer. Many different formalizations are possible e.g.: – Predicate logic or one of its fragments like terminological logic – Object-oriented representation – frame logic – ... An important role plays the choice of the vocabulary (see chapter on Knowledge Containers) Because one might obtain ontologies from different sources the problem of (automatic) translation of ontologies arises.

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Components of an Ontology (1) •

Lexical Level: – LC: A lexicon of signs (words) for concepts – LR: A lexicon of signs (words) for relations

•

Conceptual Level: – C: A set of concepts, hierarchically organized in a hierarchy H • can be considered classes in an OO class hierarchy • can also be considered as unary predicates in logics – R: A set of relations among concepts • can be considered as relational attributes • can also be considered as n-ary (n>=2) predicates in logics – A: A set of axioms over concepts and relations • logical formulas/rules stating properties of relations or constraints. • similarity measures (if they represent a shared utility estimation)

•

Mapping: – F ⊆ LC × C : A mapping relation of lexicon items to concepts – G ⊆ LR × R : A mapping relation of lexicon items to relations

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Example (1): English - German Person

Person

Firma

member

Mensch

client

Company

Mitglied *

employee

Employee

Angesteller

*

Project

+memberOf *

+client

*

*

Angestellte

Company *

+hasParticipant

memberOf(X,Y) :- hasParticipant(Y,X). Manager

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Example (2) •

Lexicon:

•

Lexicon:

•

Concepts: C = { person, employee, manager, project, company, ...} H = { manager is-a employee, employee is-a person, ... } Relations: R = { hasParticipant ⊆ project × person, memberOf ⊆ person × project, client ⊆ project × company, ...} Axioms: A = { memberOf(X,Y) :- hasParticipant (Y,X). ... }

•

• • •

LC= { “employee“, “Angestellter“, “Angestellte“, “Organisation“, “Projekt“, ... } LR = { “member”, “Mitglied”, “client”, ... }

Mapping: F = { (“employee”,employee), (“Angestellter”,employee), (“Angestellte”,employee), (“Projekt”,project), ... } Mapping: G = { ( “member”,memberOf ), ( “Mitglied”,memberOf ) , (“client”, client ), ... }

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Example Ontology in F-Logic Concept Hierarchy

Slot Definitions

Rules

Object[]. Person[ Person :: Object. firstName =>> STRING; Employee :: Person. lastName =>> STRING; Researcher :: Employe. eMail =>> STRING; Publication::Object. ... publication =>> Publication]. Employee[ affiliation =>> Organization; ...]. Researcher[ researchInterest=>>ResearchTopic; cooperatesWith=>> Researcher]. Publication[ =>> Person; author title =>> STRING; year =>> NUMBER; abstract =>> STRING].

- UoCalgary 2003 -

FORALL Person1, Person2 Person1:Researcher [cooperatesWith ->> Person2] > Person1]. FORALL Person1, Publ1 Publ1:Publication [author ->> Person1] Person1:Person [publication ->> Publ1]. FORALL O,C,A,V,T V:T >T] AND O:C[A->>V].

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Comparison of Ontologies (1) Li :: Representation language i = 1,2 Ontolologyi: Ontology in Li Expresssioni: Expression in Li

Expression1 + Ontolology1

↔ Expression2 + Ontolology2 ? How to translate it to this side ?

This side is known to me

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Comparison of Ontologies (2) • Under which conditions are two expressions in different ontologies equivalent and how can this be tested ? What is the meaning of equivalence of ontologies ? • How can one find for a given expression in one ontology some equivalent expression in another ontology (translation problem) ? • Are there universal languages to compare arbitrary ontologies (exchange formats) ? • The mapping of two expressions onto an expression in an exchange format which is equivalent to the given expressions is called semantic unification (analogous to syntactic unification defined for substitutions). - UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Semantic Unification (1) Li :: Representation language Ontolologyi: Ontology in Li Expresssioni: Expression in Li Expression1 + Ontolology1

Expression2 + Ontolology2

The same ?? Aproximate semantic unification: Almost the same ?? - UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Semantic Unification (2) • Major Problems: The languages may have – – – –

Different vocabulary Different levels of abstraction Different intentions Consequence: Introduce intermediate language levels („mediators“). – A mediator language can support • Semantic unification • Translation

• Example: Stock exchanges: – What is the value of $ in pounds in NY? – What is the $ in pounds in London?

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Reuse of Ontologies (1) • The translation makes the reuse of ontologies within the same representation language possible. • Example (for predicate logic): • Told is supposed to be an ontology for the organization of personell with a predicate vacation(space, location) (shall express: when and where) with corresponding regulations. Furthermore let Lnew be a language with the predicate holidays(time). A suitable translation is τ(holidays(x)) = ∃ y(vacation(x,y)) • The knowledge base of Told can now be reused.

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Reuse of Ontologies (2) • For different logical languages, e.g. differently represented fragments of predicate logic it is necessary to – understand the expressions of one language syntactically in the other language • or – to understand the expressions of both languages in a third language (exchange format). • The condition in order to achieve this is that the expressions are not ambigious.

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

The Big Picture Inference Engine, e.g. Logic Inference or CBR

Query Interface

Ontology

concept xyz

concept xyz relation abc ....

concept xyz relation abc ....

concept xyz relation abc ....

Document HTML

Document doc

Document pdf

Document gif

Meta Data relation abc Annotation .... (e.g. in RDF)

Human Readable

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Meta Data •

Generation of Meta Data: – Manually: tools are needed to annotate the documents with meta data using a given ontology (e.g. Tool OntoAnnotate from Ontoprise). – Automatically: tools are needed to convert documents into meta data using text analysis methods (e.g. Tool orenge:TextMiner)

•

Representation of Meta Data – XML as part of the document itself – XML as separate document containing references to the doc

no web standard for ontologies, but tool-specific formats

– RDF/RDF Schema separate resource description

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Current Players • USA: W3C, DARPA, NSF, Maryland, Stanford, ... • Canada: NRC-IIT-CISTI, ... • Europe: IST – – – – – –

• • • •

Netherlands: Amsterdam, Twente, ... UK: Manchester, Newcastle, ... France: INRIA , ... Germany: Karlsruhe, DFKI, Hannover, Bremen, IW-Köln, ... Sweden: Linköping Switzerland: MCM

Japan: INTAP, Keio, ... Korea: KAIST Australia: Melbourne, ... ...

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Summary • • • • •

General and task ontologies Terminology, glossary and thesaurus Formalization of ontologies Different ontologies Reuse

- UoCalgary 2003 -

(c) 2003 Prof. Dr. Michael M. Richter, Universität Kaiserslautern