Ontology management

1
Ontology management

• Michel Klein
• Ying Ding
• Free University, Amsterdam

2
Outline

• questions that ontology management should answer
• problems in combined use of ontologies
• some existing approaches
• summary of techniques
• plans in workpackage

3
Questions

• central aspect re-use
• how to align different domain descriptions
• how to translate ontologies
• how to build ontologies from components
• how to check the result of integration
• how to handle changes over time
• how to maintain versions of ontologies
• how to store ontologies
• how to identify and retrieve ontologies

4
Problems

• aligning and translatingmismatches between
  ontologies
• building include mechanisms, redefinition
  mechanisms
• changes and versioning impact on already
  annotated resources, relation between versions
• storage and identification description of
  content ( ontology itself!)

5
Mismatches between ontologies

• Language level
• Syntax
• Representation
• Semantics
• Expressivity

• Model level
• Modeling style
• Paradigm
• Concept description
• Conceptual
• Coverage of model
• Scope of concepts
• Terminological
• Synonyms
• Homonyms
• Encoding / multi-Language

6
Language level mismatches

• Syntax way of writing things
• Class definition
• (RDF Schema)
• defconcept Class (LOOM)
• Representation way of representing conceptual
  notions
• representing equivalence
• explicit equivalence statement or
• mutual subclass-of statements
• Semantics exact meaning notions
• multiple domain statements
• intersection (OIL RDF Schema)
• union (RDF Schema)
• Expressivity which notions can be expressed
• can negation be expressed in some way, or not
• the support of lists and sets, default values,
  quantification etc.

7
Model level mismatches

• Modeling style
• Paradigm how specific concepts are represented
• time interval logic or point representation
• different top level ontology (hierarchy)
• Concept description how a concept is described
• the way the is-a hierarchy is built
• publicationon
• publication
• are distinctions made by attributes or subclasses
• using a qualifying attribute (motorbike
  motor-vehicle with only two wheels)
• introducing a separate class (motorbike)

8
Model level mismatches (cont.)

• Conceptualization
• Model coverage and granularity which things are
  described and in how much detail
• one ontology only models cars, not trucks
• one models trucks, but only with very light
  hierarchy
• one models trucks with fine-grained (deep)
  hierarchy
• two ontologies both model trucks in detail, but
  to different user community (truck producer vs.
  truck repairer)
• Scope of concept what are the instances of a
  concept
• employee can mean all people that have a room
  within a company or all people that get paid by
  a company

9
Model level mismatches (cont.)

• Terminological
• Synonymsseveral words for same concept
• employee (HR)staff (Administration)researcher(R
  D)
• carautomobile
• Homonyms one word with several meanings
• bank river bank vs. financial bank
• fan cooling system vs. sports fan
• Encoding / Language what linguistic or metric
  system is used
• English or Dutch
• inches or centimeters
• Almost always, language mismatches coincide with
  scope mismatches

10
Tools and techniques

• Chimaera
• ontology merging and diagnosis tool (KSL,
  Stanford University)
• merging ontology
• unifying the semantically identical terms
  (T1(O1)T2(O2)T(O))
• relating terms based on subsumption,
  disjointness, or instance relationship
• name resolution listssuggesting candidates
  (classes or slots) to be merged
• taxonomy resolution lists suggesting taxonomy
  areas to be reorganized.
• diagnosis support verifying, validating and so
  on.
• Summary it solves mismatches at the
  terminological level in a very light way, finds
  editing point for the users, provides diagnostic
  function.

11
Tools and techniques

• PROMT
• interactive ontology-merging tool (Noy Musen,
  2000)
• providing suggestion list, conflict-detecting
• starting with linguistic similarity matching of
  frame names
• detecting conflicts after the user action
• proposing to-do list
• ontology merging operations
• merge classes, merge slots, merge bindings
  between slot and class,
• perform deep copy (including all the parents and
  classes, slots it refers to)
• perform shallow copy (only class itself)
• kinds of conflicts
• name conflicts, dangling references(non-existing
  references), redundancy (more than one path from
  a class to a parent), slot-value restrictions
  that violate class inheritance, etc.
• Summary iterative suggestions for concept merges
  based on linguistic and structural knowledge.

12
Tools and techniques

• OntoMorph
• a transformation system for symbolic knowledge
• mechanisms (mixture of both)
• syntactic rewriting pattern-directed rewrite
  rules based on pattern-matching
• semantic rewriting modulating syntactic
  rewriting via semantic models and logical
  inference
• facilitating ontology merging by transforming the
  ontologies into a common format with common
  names, common syntax, uniform modeling
  assumptions, etc.
• Summary solve problems at the language level of
  ontology mismatches, make it possible to specify
  the transformations both at a syntactic level and
  semantic level

13
Tools and techniques

• Others
• Ontolingua (KSL of Stanford University) a set of
  tools for analyzing and translating ontologies.
• It consists of a server and a representation
  language.
• This server provides a repository of ontologies.
• The ontology stored at the server could be
  converted into different formats (languages).
• WebOnto (Knowledge Media Institute of the Open
  University) supporting the collaborative
  browsing, creating and editing of ontologies.
• a direct manipulation interface displaying
  ontological expressions,
• an ontology discussion tool called Tadzebao
  supporting both asynchronous and synchronous
  discussions on ontologies.
• KACTUS toolkit VOID, is an interactive
  environment for browsing, editing and managing
  (libraries of) ontologies.
• organizing libraries of ontologies, translating
  between different ontology formalisms,
• browsing, editing and querying ontologies in
  various formalisms,
• reusing ontologies, handling various ontology
  formalisms (CML, EXPRESS and Ontolingua) and
  performing (partial) translations between these
  formalisms.

14
Existing approaches

• Superimposed metamodel (Bowers Delcambre, 2000)
• Objectives transforming information from one
  representation to another
• Methodology focusing on model-based information
  where information representation scheme provides
  structural modeling constructs.
• Three types of mapping
• Inter-model mapping mapping the ontology
  language constructs for ontology translation
• Inter-schema mapping defining the relation
  between ontology elements for data translation
• Model-to-schema mapping combing the above two
• Summary solving mismatches at the language
  level, the semantic resolution of mismatches at
  the ontology level is left open.

15
Existing approaches (cont.)

• Layered approach (Melnik Decker, 2000)
• Objectives facilitating data interoperation
  using a layered approach
• Methodology clearly separating different layers
  will ease the achievement of interoperability
• Three layers (each of layers has sub-layers)
• Syntax layer dumbing down object-oriented info
  into document instances and byte streams
• Object layer (frame layer) providing an
  object-oriented view of the domain
• Semantic layer (knowledge representation layer)
  conceptual modeling knowledge engineering tasks
• Summary only solving some of the language level
  mismatches (e.g., syntax, representation)

16
Existing approaches (cont.)

• OKBC (Open Knowledge Base Connectivity, Chaudhri
  et al., 1998)
• Methodology defining a mapping from OKBC
  knowledge model to the specific ontology language
• It could solve some mismatches at the language
  level
• SHOE (Ontology integration and versioning, Helfin
  and Hendler, 2000)
• Three ways for ontology integration
• OM O1O2 (inference rules are defined to map
  the common items between O1 and O2)
• Terminological differences if-and-only-if rules
• Scope differences mapping to most specific
  category based on subsumption
• Encoding differences handled by mapping
  individual values
• O1O1 OM, O2O2OM (revising each ontology to
  include mapping)
• BACKWORD-COMPATIBLE-WITH specifying revisions in
  place of the original ontology
• Interaction ontology common items of O1 and O2
• Summary SHOE allows to prove some results of
  ontology difference on the same way also preserve
  the intended meaning.

17
Existing approaches (cont.)

• Ontology algebra (Stanford)
• objective to provide the capability for
  interrogating many largely semantically disjoint
  knowledge resources.
• component intersection, union and difference.
• articulations contexts
• articulations (the rules that provides links
  across domains) enable the knowledge
  interoperability.
• Contexts (the abstract mathematical entities with
  some properties) provide guarantees about the
  knowledge they export, and contain the inferences
  feasible checking.
• results The ontology resulting from the mappings
  is assumed to be consistent only within its own
  context, called an articulation context.
• Applications
• CYC project microtheory context
• ONION (Ontology compositION) ontology is
  represented by the conceptual graph and the
  ontology mapping is based on the graph mapping.
  The main innovation
• using articulations of ontologies to interoperate
  among ontologies
• representing ontologies graphically which could
  help in separating the data layer with the
  inference engine
• separating the logical inference engine from the
  representation model as much as possible so as to
  allows accommodating different inference engines.
  
• SKAT (Semantic Knowledge Articulation Tool)
  extracting structural information from an
  ontology for creating a new ontology.
• SMART providing a semi-automatic approach to
  ontology merging and alignment. It also can pop
  out the to-do lists and do the consistency
  checking.

18
Existing approaches (cont.)

• AIFBs method
• Objectives an order-theoretic foundation for
  maintaining and merging ontologies
• Methodology Formal Concept Analysis (a
  comprehensive formalization of concepts by
  mathematizing them as a unit of thought
  constituted of two parts its extension and its
  intension)
• Future direction the integration of approaches
  for generating ontologies from text with
  linguistic methods, and for evaluating them with
  data mining techniques.
• InfoSleuths reference ontology
• Objectives supporting construction of complex
  ontologies from smaller component ontologies.
• Methodology mapping is explicitly specified
  among these ontologies as relationships between a
  term in one ontology and the related terms in
  other ontologies.
• All mappings between ontologies are made by a
  special class of agents called resource agents.
• KRAFTs ontology clustering
• one-to-one mapping class mapping, attribute
  mapping, relation mapping, compound mapping.
• Ontology clustering based on the similarities
  between the concepts known to the different
  agents.
• represented in the hierarchy fashion.
• new ontology cluster is a child ontology that
  defined certain new concepts using the concepts
  already contained in its parent ontology.
• concepts are described in terms of attributes and
  inheritance relations, and are hierarchically
  organized.
• Application international coffee preparation.

19
Existing approaches (cont.)

• Others
• Williams Tsatsoulis (2000) proposed an
  instance-based approach for identifying candidate
  relations between diverse ontologies using
  concept cluster integration. However, here they
  assume all the relations are only general is-a
  relations.
• Tamma Bench-Capon (2000) presented a
  semi-automatic framework to deal with inheritance
  conflicts and inconsistencies while integrating
  ontologies. The ontologies are represented by a
  frame-based language where the set of slot facets
  is extended to encompass other information in
  order to associate with each attribute a degree
  of strength thus permitting to deal with default
  conflicts and inconsistencies.
• Hovy, E. (1998) described several heuristics
  rules to support the merging of ontologies. For
  instance, the NAME heuristic compares the names
  of two concepts, the DEFINITION heuristics uses
  linguistic techniques for comparing the natural
  language definitions of two concepts, and the
  TAXONOMY heuristic checks the closeness of two
  concepts to each other.

20
Summary

• Aligning mapping
• Heuristic matching algorithms
• Linguistic based matching word-stem, WordNet,
  simple name matching by ignoring hyphens or
  capitalizing
• Structural and model matching language
  translation, mapping based on subsumption,
  instance relations
• Providing to-do list, consistency checking,
  conflict-handling, editing point
• Diagnosing or checking
• Domain independent verification and validation
  checks name conflicts, dangling reference
• Validation required reasoning redundancy in
  class hierarchy, value restriction
• Interoperability
• Aligning the metamodel
• Layered interoperability
• Transformation rules
• Mapping to the common knowledge model
• Versioning
• OM O1O2
• O1O1 OM, O2O2OM
• Interaction ontology common items of O1 and O2

21
Plans

• Structure of Ontology Management Del
• Survey existing approaches
• Ontology library requirements, save, retrieve,
  reuse
• Ontology mapping methods examples
• Ontology versioning framework, methodology