Transcript a = b - Centre LGI2P

A
[Treemaps]
O
I
C
Application
Context 1
T
F
R
D
I
F
[Touch-Graph]
[ToxNuc-e project]
Application
Context 2
[MBox Project]
[Kartoo]
Ontological Distance Measures for Information
Visualisation on Conceptual Maps
Sylvie Ranwez
Jean Villerd
Michel Crampes
LGI2P Research Centre – EMA, Nîmes
Vincent Ranwez
ISEM – Montpellier University
Overview
 Semantic distances: state-of-the-Art
 From ontology to semantic distance
•
•
•
•
Intuitive approach
Formal definition
Example
Distance properties
 Resulting visualisation
 Discussion and perspectives
 Conclusion
Ontological Distance Measure for Information Visualisation on Conceptual Maps - S. Ranwez
3
Semantic distances: state-of-the-Art
Estimating similarity between concepts
Methods based on the concept hierarchy
 d(a, b): the length of the shortest path between a and b [Sowa]
 sim(a, b): function of common subsumers [Resnik]
Considers only one point of view on the concept
Supposes homogeneity of branches’ semantic
Does not respect distances properties
Methods based on vectors calculus




Complementarity
of the two
approaches
Vectors of terms to describe a document
Vectors of concepts to describe a given concept
Ensemblist methods (Dice or Jaccard)
Geometric methods (cosines), Euclidian measure, distributional, etc.
Vectors are not always available
Lack of precision due to the vectorisation (synonyms)
Ontological Distance Measure for Information Visualisation on Conceptual Maps - S. Ranwez
4
Overview
 Semantic distances: state-of-the-Art
 From ontology to semantic distance
•
•
•
•
Intuitive approach
Formal definition
Example
Distance properties
 Resulting visualisation
 Discussion and perspectives
 Conclusion
Ontological Distance Measure for Information Visualisation on Conceptual Maps - S. Ranwez
5
From ontology to semantic distance
 Intuitive approach on the is-a relation
Two concepts are close if there is a concept that sumbsumes both of them and
if this concept is slightly more general (encompasses few more concepts)
T
[MeSH]
Persons (44)
Occupational Groups (12)
…
Health Personnel (20)
Dentists (1) …
Veterinarians (0)
Administrative Personnel (4)
Nurses (6)
d(Veterinarians, Nurses) < d(Trustees, Nurses)
Physician Executives (0) …
Trustees (0)
d(Nurses, Health Personnel) < d(Veterinarians, Health Personnel)
Ontological Distance Measure for Information Visualisation on Conceptual Maps - S. Ranwez
6
From ontology to semantic distance
 Intuitive approach on the is-a relation
However multiple inheritance (points of view) must be taken into account
T
Persons (44)
Occupational Groups (12)
…
Health Personnel (20)
Dentists (1) …
Veterinarians (0)
Administrative Personnel (4)
Nurses (6)
Nurses Administrators (0) Physician Executives (0) … Trustees (0)
Ontological Distance Measure for Information Visualisation on Conceptual Maps - S. Ranwez
7
From ontology to semantic distance
 Definition
T
C0
C1
C4
a
C9
C2
C5
C3
C6
C10
C7
b
C8
C11
ancExc(a,b)
) desc(a)
desc(a)
desc( ancExc(a,b)
)

desc(b)
- desc(a)
 desc(b)
dISA(a, b) = |desc(
desc(
ancExc(a, b)

desc(a)
desc(b)
desc(b)
- desc(a)
 desc(b) |
dISA(a, b) = 11
Ontological Distance Measure for Information Visualisation on Conceptual Maps - S. Ranwez
8
From ontology to semantic distance
dISA(a, b) = | desc( ancExc(a, b) )  desc(a)  desc(b) - desc(a)  desc(b) |
 Example
…
Persons (44)
Occupational Groups (12)
…
Health Personnel (20)
Dentists (1) …
Veterinarians (0)
Administrative Personnel (4)
Nurses (6)
Nurses Administrators (0)
Physician Executives (0) … Trustees (0)
dISA(Trust., Nur.) = | desc( ancExc(Trust., Nur.)  desc(Nur.)  desc(Trust.) - desc(Nur.)  desc(Trust.) |
dISA(Trust., Nur.) = | desc(Health P., Admin P.)  {Nur., …, Nur. adm.}  {Trust.} -  |
dISA(Trust., Nur.) = | {Health P., Dentists, …, Nur., Nur. adm., Admin P., …, Trust.} | = 59
dISA(Nur. adm., Phys. Exec.) = 8
dISA(Trust., Phys. Exec.) = 58
Ontological Distance Measure for Information Visualisation on Conceptual Maps - S. Ranwez
dISA(Nur., Phys. Exec.) = 13
9
From ontology to semantic distance
dISA(a, b) = | desc( ancExc(a, b) )  desc(a)  desc(b) - desc(a)  desc(b) |
 Respects the three properties of a distance
• Positiveness :  a, b
dISA(a, b)  0 and dISA(a, b) = 0  a = b
• Symmetry :  a, b
dISA(a, b) = dISA(b, a)
• Triangle inequality :  a, b, c
dISA(a, c) + dISA(c, b)  dISA(a, b)
 Extension
• Intuitive distance in a tree-like hierarchy when a subsumes b
dISA(a, b) = | desc(a) – desc(b) |
Ontological Distance Measure for Information Visualisation on Conceptual Maps - S. Ranwez
10
Overview
 Semantic distances: state-of-the-Art
 From ontology to semantic distance
• Intuitive approach
• Formal definition
• Example
 Resulting visualisation
 Discussion and perspectives
 Conclusion
Ontological Distance Measure for Information Visualisation on Conceptual Maps - S. Ranwez
11
Resulting visualisation
dISA(Trust., Nur.) = 59
dISA(Nur. adm., Phys. Exec.) = 8
dISA(Trust., Phys. Exec.) = 58
dISA(Nur., Phys. Exec.) = 13
…
Persons (44)
Occupational Groups (12)
Health Personnel (20)
…
Administrative Personnel (4)
Dentists (1) … Veterinarians (0) Nurses (6)
Nurses Administrators (0) …
Ontological Distance Measure for Information Visualisation on Conceptual Maps - S. Ranwez
Trustees (0)
12
Resulting visualisation
Nervous System
Diseases
Example from the MeSH
Central Nervous
System Diseases
Neurologic
Manifestations
Brain Diseases
Headache Disorder
Pathological
Conditions,
Signs and
Symptoms
Sign and Symptoms
Pain
Headache Disorder,
Primary
Headache
…
Migraine = Migraine
Disorder
Migraine
Disorder
with Aura
Migraine
Disorder
without
Aura
Ontological Distance Measure for Information Visualisation on Conceptual Maps - S. Ranwez
13
Discussion and perspectives
Towards a semantic distance



Combine the ISA distance with other distance measures taking into
account other kinds of relations
Combine with approaches using vector calculus
Combine the ISA distance with the level of detail of the concepts
Validation and extension of the visualisation
1. Visualisation of ontologies by projection and identification of clusters
2. Use of traditional clustering methods (hierarchical clustering, K-means…)
3. Comparisons and validation of our approach
Enforce the use in industrial context



Validation of existing ontologies
Support during the conception of new ontologies
Support while navigating or searching for information
Ontological Distance Measure for Information Visualisation on Conceptual Maps - S. Ranwez
14
Conclusion
 Proposition of a distance using ISA relations, that respects the
distance properties
• Positiveness
• Symmetry
• Triangle inequality
 Projection of ontologies: a new way of visualising ontologies
• Towards conceptual maps
• Support in ontologies building and validating
 Application
• Ontology design
• Navigation support
• Information retrieval
Ontological Distance Measure for Information Visualisation on Conceptual Maps - S. Ranwez
15
Ontological Distance Measures for Information
Visualisation on Conceptual Maps
[email protected]
http://www.lgi2p.ema.fr/~ranwezs
[email protected]
http://ranwez.free.fr/
[email protected]
http://www.lgi2p.ema.fr/~villerd
[email protected]
http://www.ema.fr/~mcrampes