Association%20in%20Level%202%20Fusion

1
Association in Level 2 Fusion

• Mieczyslaw M. Kokar
• Christopher J. Matheus
• Jerzy A. Letkowski
• Kenneth Baclawski
• Paul Kogut

2
Overview

• Data association in Level 1 and Level 2
• Ontologies
• Reasoning process
• Examples of reasoning about associations
• Useful OWL constructs
• Confidence of association
• Conclusion

3
Level 1 Data Association

• Data to object
• Measurement to object
• Measurement to track
• Object to object
• Track to track
• ID to ID
• Approach
• Define a measure of distance (quantitative)
• Minimize distance

4
Level 2 Data Association Problem

• Data may be text
• Identification of the same individuals called
  co-reference
• No quantitative properties of objects
• Text processing tools used
• Possibly translated to logic
• Logic based tools (reasoners or theorem provers)
  are used for deriving conclusions
• Why not for data association?

5
Ontology

• An explicit specification of a conceptualization
  the objects, concepts, and other entities that
  are assumed to exist in some area of interest and
  the relationships that hold among them
  (Genesereth Nilsson, 1997)
• Definitions associate the names of entities in
  the universe of discourse (e.g., classes,
  relations, functions, or other objects) with
  human-readable text describing what the names
  mean, and formal axioms that constrain the
  interpretation and well-formed use of these
  terms. A statement of a logical theory. (Gruber)
• An agent commits to an ontology if its observable
  actions are consistent with the definitions in
  the ontology (knowledge level).
• A common ontology defines the vocabulary with
  which queries and assertions are exchanged among
  agents.

6
An Ontology (in UML)
7
Identifying Associations in Level 2 Fusion
8
Ontology Fragment in OWL
ltowlClass rdfID"Person"gt ltrdfslabel
xmllang"en"gtPersonlt/rdfslabelgt
ltrdfssubClassOfgt ltowlRestrictiongt
ltowlonProperty rdfresource"personName"/gt
ltowlcardinality rdfdatatype"http//www.w3.or
g/2000/10/XMLSchemanonNegativeInteger"gt1lt/owlcar
dinalitygt lt/owlRestrictiongt
lt/rdfssubClassOfgt ltrdfssubClassOfgt
ltowlRestrictiongt ltowlonProperty
rdfresource"personAge"/gt
ltowlcardinality rdfdatatype"http//www.w3.org/2
000/10/XMLSchemanonNegativeInteger"gt1lt/owlcardin
alitygt lt/owlRestrictiongt
lt/rdfssubClassOfgt lt/owlClassgt ltowlClass
rdfID"Leader"gt ltrdfslabel
xmllang"en"gtLeaderlt/rdfslabelgt
ltrdfssubClassOf rdfresource"Person"/gt
ltrdfssubClassOfgt ltowlRestrictiongt
ltowlonProperty rdfresource"leaderOf"/gt
ltowlcardinality rdfdatatype"http//www.w3.org/
2000/10/XMLSchemanonNegativeInteger"gt1lt/owlcardi
nalitygt lt/owlRestrictiongt
lt/rdfssubClassOfgt lt/owlClassgt ltowlClass
rdfID"Organization"gt ltrdfslabel
xmllang"en"gtOrganizationlt/rdfslabelgt
ltrdfssubClassOfgt ltowlRestrictiongt
ltowlonProperty rdfresource"ledBy"/gt
ltowlcardinality rdfdatatype"http//www.w3.org/2
000/10/XMLSchemanonNegativeInteger"gt1lt/owlcardin
alitygt lt/owlRestrictiongt
lt/rdfssubClassOfgt lt/owlClassgt
9
Ontology (cont.)
ltowlObjectProperty rdfID"personName"gt
ltrdfsdomain rdfresource"Person" /gt
ltrdfsrange rdfresource"http//www.w3.org/2000/1
0/XMLSchemastring" /gt lt/owlObjectPropertygt ltowl
ObjectProperty rdfID"personAge"gt
ltrdfsdomain rdfresource"Person" /gt
ltrdfsrange rdfresource"http//www.w3.org/2000/1
0/XMLSchemastring" /gt lt/owlObjectPropertygt
ltowlObjectProperty rdfID"orgName"gt
ltrdfsdomain rdfresource"Organization" /gt
ltrdfsrange rdfresource"http//www.w3.org/2000/1
0/XMLSchemastring" /gt lt/owlObjectPropertygt ltowl
ObjectProperty rdfID"orgType"gt
ltrdfsdomain rdfresource"Organization" /gt
ltrdfsrange rdfresource"http//www.w3.org/2000/1
0/XMLSchemastring" /gt lt/owlObjectPropertygt ltowl
ObjectProperty rdfID"leaderOf"gt
ltrdfsdomain rdfresource"Leader" /gt
ltrdfsrange rdfresource"Organization" /gt
lt/owlObjectPropertygt ltowlObjectProperty
rdfID"ledBy"gt ltowlinverseOf
rdfresource"leaderOf" /gt lt/owlObjectPropertygt
10
Annotation
ltLeader rdfID"A1"gt ltpersonName
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gtOsamalt/personNamegt ltpersonAge
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gtoldlt/personAgegt ltleaderOf
rdfresource"X"/gt lt/Leadergt ltLeader
rdfID"A2"gt ltpersonName rdfdatatype"http//
www.w3.org/2000/10/XMLSchemastring"gtOsamalt/person
Namegt ltpersonAge rdfdatatype"http//www.w3.o
rg/2000/10/XMLSchemastring"gtoldlt/personAgegt
ltleaderOf rdfresource"X"/gt ltowlsameAs
rdfresource"A1"/gt lt/Leadergt ltLeader
rdfID"B1"gt ltpersonName rdfdatatype"http//
www.w3.org/2000/10/XMLSchemastring"gtStrossenlt/per
sonNamegt ltpersonAge rdfdatatype"http//www.w
3.org/2000/10/XMLSchemastring"gtmaturelt/personAgegt
ltleaderOf rdfresource"Y"/gt lt/Leadergt
ltOrganization rdfID"X"gt ltorgName
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gtal-Quedalt/orgNamegt ltorgType
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gtterroristlt/orgTypegt ltledBy
rdfresource"A1"/gt ltheadquarters
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gtUnknownlt/headquartersgt lt/Organizationgt ltOr
ganization rdfID"Y"gt ltorgName
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gtACLUlt/orgNamegt ltorgType
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gtnon-profitlt/orgTypegt ltledBy
rdfresource"B1"/gt ltheadquarters
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gtUSAlt/headquartersgt lt/Organizationgt
11
Association Through Reasoning

• The OWL language specifies names of individuals
  using a Uniform Resource Identifier (URI).
• Characteristics of individuals (such as position
  or velocity) are defined using properties.
• An OWL property that specifies a characteristic
  of an individual is called a DatatypeProperty.
• Relations between individuals are also specified
  using properties. An OWL property of this kind
  is called an ObjectProperty.
• There are two special OWL properties that can be
  used for explicit data association sameAs and
  differentFrom.
• A whole set of URIs can refer to different
  individuals by using the AllDifferent construct.
• OWL properties are binary.
• OWL properties are many-to-many unless one
  specifies constraints on the properties.

12
Use of ConsVISor(a rule-based consistency
verification tool)http//vistology.com/consvisor
Ground Truth Two names represent the same
individual and the ontology has support for
this. Two names represent two different
individuals and the ontology has support for this.

• Hypotheses (statements added to an annotation)
• The names represent the same individual in the
  world (expressed in OWL as sameAs)
• The names represent different individuals in the
  world (expressed in OWL as differentFrom)

ConsVISors Decision ConsVISor can decide that
a given annotation (with the added hypothesis) is
either consistent or inconsistent .
13
Association and Consistency

• Consider a case where ConsVISor returns
  consistent for both hypotheses.
• Since it is consistent to believe either that
  they are the same or that they are different we
  cannot make any association claims at all (not
  enough information in the annotations)
• The converse case in which both hypotheses
  produce inconsistent results should never occur
  unless there is an inconsistency in the
  underlying ontology.
• The interesting cases occur when one hypothesis
  is consistent and the other is inconsistent.
• If the sameAs hypothesis is consistent and the
  differentFrom inconsistent, they must be
  co-references.
• Conversely, if the sameAs hypothesis is
  inconsistent and the differentFrom hypothesis is
  consistent, the two references cannot refer to
  the same individual.

14
Ground Truth
Hypothesis
ConsVISors Decision
Annotation Example
http//vistology.com/exp/1.owl
?
consistent
inconsistent
?
ltsameAsgt
Impossible!
http//vistology.com/exp/3.owl
?
consistent
Two individuals are the same
ltdifferentFromgt
?
http//vistology.com/exp/4.owl
inconsistent
http//vistology.com/exp/5.owl
?
consistent
Two individuals are different
ltsameAsgt
inconsistent
?
http//vistology.com/exp/6.owl
http//vistology.com/exp/7.owl
?
ltdifferentFromgt
consistent
inconsistent
Impossible!
?
15
Reasoning about Associations
ltLeader rdfID"A1"gt ltpersonName
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gtOsamalt/personNamegt ltpersonAge
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gtoldlt/personAgegt ltleaderOf
rdfresource"X"/gt lt/Leadergt ltLeader
rdfID"B1"gt ltpersonName rdfdatatype"http//
www.w3.org/2000/10/XMLSchemastring"gtStrossenlt/per
sonNamegt ltpersonAge rdfdatatype"http//www.w
3.org/2000/10/XMLSchemastring"gtmaturelt/personAgegt
ltleaderOf rdfresource"Y"/gt
ltowlsameAs rdfresource"A1"/gt lt/Leadergt
ltOrganization rdfID"X"gt ltorgName
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gt al-Quedalt/orgNamegt ltorgType
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gtterroristlt/orgTypegt ltledBy
rdfresource"A1"/gt lt/Organizationgt
ltOrganization rdfID"Y"gt ltorgName
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gtACLUlt/orgNamegt ltorgType
rdfdatatype"http//www.w3.org/2000/10/XMLSchema
string"gtnon-profitlt/orgTypegt ltledBy
rdfresource"B1"/gt lt/Organizationgt
ConsVISors Decision inconsistent
16
Useful OWL Constructs

• FunctionalProperty and InverseFunctionalProperty
• If a property is functional then a given
  individual can be related to at most one other
  value or individual.
• If a property is inverse functional, then an
  individual or value can be used by at most one
  individual.
• For example, one specifies that a leadership is
  functional by asserting that ledBy is an instance
  of FunctionalProperty or that its inverse
  leaderOf is an instance of InverseFunctionalProper
  ty. If one of these is specified and if two
  individual names are both known to refer to the
  leader of the same organization, then those names
  must represent the same individual.
• Cardinality constraints.
• A cardinality constraint restricts the number of
  individuals (or values) that a given individual
  can be related to.
• A maxCardinality constraint specifies an upper
  bound on this number, a minCardinality constraint
  specifies a lower bound.
• A cardinality constraint specifies the exact
  number.
• The maxCardinality and cardinality constraints
  are especially useful for proving that two
  individual names refer to the same individual.

17
Useful OWL Constructs Cont.

• disjointWith
• When two classes are specified to be disjoint,
  then they cannot have any instances (individuals)
  in common.
• complementOf
• an extreme form of disjointness
• oneOf
• members of a class must all come from a given
  list of instances.
• unionOf
• a class is the union of some other classes.
• subClassOf, equivalentClass
• subPropertyOf, equivalentProperty
• domain and range
• SymmetricProperty and TransitiveProperty

18
Confidence

• No notion of uncertainty in OWL (so far)
• OWL reasoners dont have any way to deal with
  uncertainty even if an ontology has an explicit
  way for incorporating uncertainty into its
  annotations
• ConsVISor can provide annotations involved in the
  determination of an inconsistency (trace)
• Using an external process it is possible in some
  cases to combine the uncertainties of the
  implicated annotations to make statements about
  the uncertainty associated with an association
  claim
• E.g., map ontology/annotation into a Bayesian net
• Work in progress

19
Conclusion

• Showed examples of reasoning about co-reference
  in annotations (in OWL)
• Merge two annotation files
• Insert sameAs or differentFrom hypothesis
• Use ConsVISor consistency checker
• Consistent - plausible
• Inconsistency evidence against
• OWL becoming a de-facto standard
• Discussed useful OWL constructs
• Confidence work in progress (Bayesian nets)
• Using ConsVISor in a system for generating and
  testing association hypotheses future work