An Algorithm and Implementation for GeoOntologies Integration

1
An Algorithm and Implementation for GeoOntologies
Integration

• Guillermo Nudelman Hess
• Prof. Dr. Cirano Iochpe
• Prof.ssa. Dr. Silvana Castano

2
Agenda

• Introduction
• Motivating example
• G-Match
• Extension of conventional matcher
• Complete new matcher
• Results
• Conclusions and future directions

3
Introduction

• Current scenario
• High cost of geographic data acquisition
• Use of geographic ontologies getting popular
• GML in OWL
• ISO 19109
• Academical initiatives
• Web enables interchange of information

4
Introduction

• Current efforts and challenges
• Interoperability / integration
• Semantics

• HOW TO DETERMINE IF TWO GEOGRAPHIC ELEMENTS
  (CONCEPTS, CLASSES) ARE EQUIVALENT?

5
Introduction

• Solution
• Ontology matching ? Similarity measurement
• Ontology matching today
• Conventional matcher do not consider geospatial
  particularities
• Geographic matchers are driven by geometric and
  positional features
• G-Match proposal
• Combine both conventional and spatial features in
  a unique matcher

6
Motivating example
7
Motivating example
8
G-Match

• Conventional features
• Concept names
• Attributes names
• Taxonomy
• Conventional relationships
• Spatial features
• Topological relationships

9
G-Match

• Extension of a conventional matcher
• Names, attributes and taxonomy similarity
  measured by the conventional matcher
• Conventional and topological similarity measured
  by the geographic matcher

10
G-Match

• Complete new matcher
• Names, attributes and taxonomy, conventional and
  topological similarity measured by the geographic
  matcher

11
G-Match (complete)
12
G-Match

• Two-step algorithm
• 1st similarity of names and attributes
• 2nd similarity os relationships and taxonomy
• Name similarity based on equality, liguistic
  affinity (WordNet) and string metrics
• The final similarity measure is balanced
• Each feature receives a weight
• Final value within 0,1

13
Results(from the papers example)
Matcher EM AM CAM Precision Recall
Prompt 15 14 13 93 87
H-Match 15 12 10 83 67
G-MatchE 15 13 11 85 73
G-MatchC 15 15 14 93 93
G-MatchE was implemented as an extension o H-Match
14
Conclusions

• Need of a matcher special tailored for geographic
  ontologies
• Combination of spatial and conventional features
  improve results ? balanced sum

15
Future work

• To consider also directional relations in the
  spatial similarity
• Extend G-Match to instances
• Try to (semi-)automate the tuning of weights

16
Thank you

• Guillermo Nudelman Hess
• hess_at_inf.ufrgs.br

?