Vipul Kashyap and Alex Borgida - PowerPoint PPT Presentation

1 / 22
About This Presentation
Title:

Vipul Kashyap and Alex Borgida

Description:

Concepts are collections of instances ... has been explicitly endorsed by the designers of the UMLS Semantic Network ... is a collection of kinds of infection ... – PowerPoint PPT presentation

Number of Views:57
Avg rating:3.0/5.0
Slides: 23
Provided by: vipulk
Category:
Tags: alex | borgida | kashyap | vipul

less

Transcript and Presenter's Notes

Title: Vipul Kashyap and Alex Borgida


1

Representing the UMLS Semantic Network using
OWL (Or Whats in a Semantic Web Link?)
  • Vipul Kashyap and Alex Borgida
  • NLM/NIH and Dept of CS, Rutgers University
  • kashyap_at_nlm.nih.gov, borgida_at_cs.rutgers.edu
  • The Second International Semantic Web Conference
    (ISWC)
  • Sanibel Island, Florida, October 21, 2003.

2
Outline
  • The UMLS Semantic Network
  • Vanilla Representation using OWL
  • Whats in a Semantic Network/Web Link?
  • Multiple Interpretations and Encodings of a link
  • Requirements of the UMLS? Semantic Network
  • Un-anticipated results !
  • Methodology for encoding Knowledge on the
    Semantic Web
  • First Steps
  • Conclusions and Future Work

3
The UMLS Semantic Network (partial)
4
Vanilla Representation using OWL
  • Semantic Types ? OWL classes
  • Fungus ? Organism
  • Virus ? Organism
  • Semantic Relationships ? OWL properties
  • part_of ? physically_related_to
  • contains ? physically_related_to
  • Domains and Ranges of Relationships Later
  • Properties of Semantic Network Relationships
  • Asymmetric relationships
  • has_part part_of?
  • Symmetric relationships
  • adjacent_to adjacent_to?

5
Whats in a Semantic Network/Web Link?
causes
Bacteria
Infection
  • Consider two sets
  • ?(causes) x ? Bacteria ? (?y)(y ? Infection ?
    causes(x,y))
  • Closely related to the domain of a property
  • ?(causes) ?causes.T (DL notation)
  • ?(causes) y ? Infection ? (?x)(x ? Bacteria ?
    causes(x,y))
  • Closely related to the range of a property
  • ?(causes) ?causes?.T (DL notation)

6
Interpretation 1 ?/? equals
  • ?(causes) Bacteria
  • ?(causes) Infection
  • All Bacteria have to cause and all Infections
    have to be-caused
  • No others can participate in causes

?(causes) ?(causes) b1
i1 b2 i2 b3
i3 b4 i3
Bacteria b1 b2 b3 b4
7
Interpretation 2 ?/? subsumed
  • ?(causes) ? Bacteria
  • ?(causes) ? Infection
  • Equivalent to the RDF/DAMLOIL definition of
    domain and range!
  • Not all bacteria need to cause not all
    infections have to be-caused
  • However no others can participate!

?(causes) ?(causes) b1
i1 b2 i2 b3
i2
Bacteria b1 b2 b3 b4
8
Interpretation 3 ?/? subsumes
  • ?(causes) ? Bacteria
  • ?(causes) ? Infection
  • All bacterias have to cause and all infections
    have to be-caused, but
  • A bacteria can cause a non-infection as well!
  • A non-bacteria can cause an infection as well!
  • A non-bacteria can cause a non-infection as
    well!

?(causes) ?(causes) b1
i1 b2 i2 b3
i2 b4 X
Y i3 Z T
Bacteria b1 b2 b3 b4
9
Interpretation 4 All/Some
  • All bacteria cause some infection
  • Bacteria ? ?causes.Infection
  • All bacteria must cause some infection, but
  • A bacteria can cause a non-infection as well!
  • A non-bacteria can cause an infection as well!
  • A non-bacteria can cause a non-infection as
    well!

?(causes) ?(causes) b1
i1 b2 i2 b3
i2 b4 i1
b4 X Y
i3 Z U
Bacteria b1 b2 b3 b4
10
Interpretation 5 All/Only
  • All bacteria cause only infections
  • Bacteria ? ?causes.Infection
  • All bacteria, if they cause, can cause only
    infections, but
  • Not all bacteria have to participate in the
    causes relationship
  • None of the bacteria may participate in causes
  • A non-bacteria can still cause an infection!
  • A non-bacteria can still cause a non-infection!

?(causes) ?(causes) b1
i1 b2 i2 b3
i2 Y i3 Z
U
Bacteria b1 b2 b3 b4
11
Interpretation 6 All/Each
  • All bacteria cause each infection
  • Bacteria ? ??causes.?Infection (DL notation)
  • Similar to a cross product, but
  • A bacteria can still cause a non-infection!
  • A non-bacteria can still cause an infection!
  • A non-bacteria can still cause a non-infection!

?(causes) ?(causes) bj
ik ... b1
T Y i3 Z
U
Bacteria b1 b2 b3 b4
1 ? k ? 3,
1 ? j ? 4
12
Interpretation 7 Some/Some
  • Some bacteria cause some infection
  • ? 1 (Bacteria ? ?causes.Infection) (DL notation)
  • There is at least one bacteria that causes at
    least one infection, but
  • A bacteria can still cause a non-infection!
  • A non-bacteria can still cause an infection!
  • A non-bacteria can still cause a non-infection!

?(causes) ?(causes) b2
i3 b1 T Y
i3 Z U
Bacteria b1 b2 b3 b4
13
Interpretation 8 Some/Each
  • Some bacteria cause each infection
  • ? 1 (Bacteria ? ??causes.?Infection) (DL
    notation)
  • There is at least one bacteria that causes all
    infections, but
  • A bacteria can still cause a non-infection!
  • A non-bacteria can still cause an infection!
  • A non-bacteria can still cause a non-infection!

?(causes) ?(causes) b1
ik ... b1
T Y i3 Z
U
Bacteria b1 b2 b3 b4
1 ? k ? 3
14
Requirements of the UMLS? Semantic Network
  • ? inheritance
  • If GoodBacteria and BadBacteria are
    subclasses of Bacteria
  • causes(GoodBacteria, Infection) holds
  • causes(BadBacteria, Infection) holds
  • ? inheritance (Unusual !!!!)
  • If VirulentInfection and MildInfection are
    subclasses of Infection
  • causes(Bacteria, VirulentInfection) holds
  • causes(Bacteria, MildInfection) holds
  • Inheritance blocking
  • causes(GoodBacteria, Infection) is an undesirable
    inference and should be blocked!
  • Ad-hoc polymorphism
  • causes(Bacteria, Illness)
  • causes(Diabetes, Blindness)
  • Multiple domains and ranges!

15
Unanticipated Result !
We did not anticipate the interpretation All
bacteria causes each infection to best satisfy
all the UMLS? Semantic Network Requirements
!! Please consult paper for the details of the
analysis
16
Methodology for encoding Knowledge on the
Semantic Web First Steps
Choice of Encoding
Support for Inferences?
Support for Intended Application?
Is the domain model Reasonable?
Representation/Inference in Ontology Language
Intuitive Encodings?
Inconsistency Detection?
Unintended Inferences?
Expressivity of the Ontology Language
Alternative Interpretations?
Absence of Links?
Limited Inferences?
Complexity of Operators?
Asymmetricity? Defaults?
Acceptable Approximations?
Disjoint Concepts?
In theory, there is no difference between theory
and practice. In practice, there is. -
Anonymous, Ph.D.
17
Support for Inferences of the Original Notation
  • Which notation provides a mechanism for
    inferences informally sanctioned by the notation?
  • All/Each and ?/? subsumes are the two best
    encodings
  • Does an encoding entail unintended inferences?
  • The Some/some interpretation enables upward
    inheritance of links
  • The causes property also holds for all
    superclasses of Bacteria and Infection
  • Can/should something be inferred from absence of
    a link?
  • A ? ?P.B doesnt prohibit ?A from being related
    to B
  • Should relationships be inferred to be asymmetric
    by default?
  • Automatically assert ?(P ? P?)
  • Should the is-a children be disjoint?
  • Untangling of ontologies

18
Support for Intended Application
  • Is it important to detect inconsistency in the
    ontology?
  • What is an inconsistency?
  • Are P(A, B) and P(A, C) inconsistent if B ? ?C ?
  • Will the encoding detect these inconsistencies?
  • A ? ?P.B and A ? ?P.C will not detect
    inconsistency unless we assert A ? ?P.T

19
Reasonableness of the Domain Model
  • Concepts are collections of instances
  • causes(Bacteria, Infection) relates a bacterium
    to a case of infection.
  • What are the intutive encodings?
  • All/Some and All/Only are the intuitive encodings
    used by medical ontologies based on DLs
  • All/Each and Some/Some have been rejected
  • Are alternative interpretations possible?
  • BUT The All/Each encoding satisfies all the
    UMLS? Semantic Network requirements!
  • The Some/Some encoding has been explicitly
    endorsed by the designers of the UMLS? Semantic
    Network
  • The concept INFECTION is a collection of kinds of
    infection as opposed to a collection of cases
    (instances) of infections
  • This approach has been very successful in Medical
    Information Retrieval and should be investigated
    by the Semantic Web community!

20
Representation and Inference in the Ontology
Language
  • Limitations of OWL
  • Role negation and disjunction
  • Concept cardinality
  • Can we use less expensive constructs
  • Whole body of research in complexity of DL
    operators
  • Can we simulate the desired encoding based on
    the underlying reasoner technology?
  • ? 1 C can be asserted by inserting a dummy
    instance
  • A ? ?P.B is reasoned with in a much more
    efficient way than ?P?.A ? B
  • Are there approximations to concepts and axioms?
  • Assert P1 ? P, P2 ? P instead of P ? P1 ? P2

21
Conclusions and Future Work
  • Experiences in representing a real world
    ontology, the UMLS? Semantic Network
  • Has been used very successfully
  • Requirements ?/? inheritance, inheritance
    blocking, polymorphic relationships
  • Presented multiple interpretations and encodings
    and evaluated their support for the UMLS?
    Semantic Network requirements
  • Ontology developers and encoders on the Semantic
    Web might encounter similar requirements and
    possible encodings
  • Identified criteria for choosing between the
    various encodings
  • First steps towards a methodology which might be
    useful to ontology developers
  • Ongoing and Future Work
  • Semantic Vocabulary Interoperation Project
  • http//cgsb2.nlm.nih.gov/kashyap/projects/SVIP
  • Use of OWL, RDF for improvement in Medical
    Information Retrieval

22
The formulation of a problem is often more
essential than its solution, which may be merely
a matter of mathematical or experimental skill.
To raise new questions, new possibilities, to
regard old problems from a new angle, requires
creative imagination and marks real advance in
science. - Albert Einstein, 1938.
Write a Comment
User Comments (0)
About PowerShow.com