Semantic Interoperability in GIS

1
Semantic Interoperability in GIS

• N. L. Sarda
• Suman Somavarapu

2
The Need for Interoperability

• Many sources are collecting and storing
  geographic data.
• Effective utilization only if data is sharable
  and interoperable.
• Existing Solution
• Open standards for interoperation of GIS.
• GML - Data interchange standard for geographic
  data.
• WFS, WMS - Geospatial Web Service interfaces.
• Does not address Semantic heterogeneities.

3
The Semantics Problem

• Different entities names same meaning.
• Same entities name different meaning.
• Different classification hierarchies.

4
Our Solution - SANGAM

• SemANtic Geographic data Access Methodology
• Use Ontologies to capture the semantics of the
  information and to make the content explicit
• Identify and associate semantically corresponding
  information concepts. (Ontology Mapping)
• Check consistency of the mappings using ontology
  reasoners. (Mapping Validation).

5
Our Solution - SANGAM

• Use the Ontology Mappings to generate XQuery
  transformations for data conversion. (Wrapper
  Generation).
• Accept data requests, transform them according to
  sources.
• Apply wrappers on the data got from sources and
  return merged data (Data Extraction).

6
Outline

• Ontologies and Ontology Mapping
• Ontology Reasoning
• Wrapper Generation
• Data Extraction
• Results and Analysis

7
Ontologies

• An explicit specification of a shared
  conceptualization.
• Defined using the concepts of
• Classes
• Data and Object Properties
• Constraints and Axioms

8
Geographic Ontology Model
9
Ontology Mapping

• Reconciling differences between heterogeneous
  ontologies
• Class Level Equivalent, Subclass, Superclass etc
• Eg Student gt Graduate Student
• Data Properties Equivalence, Aggregation,
  Concatenation, Transformation etc
• Eg RoadLength(in Km) RoadLength(in m)1000.
• Name ? Firstname . Lastname
• Object Properties Equivalence, Union,
  Categorization

10
Mapping Process

• Generation of lexical similarity
• Parts of speech tagging
• Semantic similarity between two words
• Semantic similarity between two descriptions
• Propagation of superclass and attribute
  similarity
• Propagation of subclass similarity
• Similar mappings.

11
Generation of lexical similarity

• WordNet - Semantic relations
• A lexical reference system.
• Nouns, verbs, adjectives and adverbs are
  organized into synonym sets.
• Each synonym set or synset represents one
  underlying lexical concept.
• Synsets are connected to one another through the
  explicit semantic relations.

12
User Interaction in Ontology Mapping
13
Layer Level Mapping
14
Attribute Level Define Transformations
15
Mapping Validation

• 1. Convert Ontology to OWL using XSLT
• 2. Use JENA Model Factory to generate a model
  instance
• 3. Set an external OWL reasoner (Pellet)
• 4. Provides inconsistencies as output.

16
OWL and Jena

• OWL- Web Ontology Language standard for
  representing
• ontologies in machine readable form.
• Constructs of classes and properties like Class,
  subClassOf,
• Property, subPropertyOf, equivalentClass,
  equivalentProperty
• Jena is a Java framework for building Semantic
  Web applications.
• Provides a programmatic environment for RDF, RDFS
  and OWL.
• Includes a rule-based inference engine. External
  reasoners can be plugged in.

17
Converting the Ontologies to OWL

• 1. Create owlOntology element with the rdfLabel
  as Source name.
• ltowlOntology rdfabout""gt
• ltrdfslabelgtTransportationlt/rdfslabelgt
• lt/owlOntologygt
• 2. For each class which doesnt have any parent
• ltowlClass rdfID"Road"/gt
• 3. For each class which has a parent
• ltowlClass rdfID"RoadPoint"gt
• ltrdfssubClassOf rdfresource"TransportationPoint
  "/gt
• lt/owlClassgt

18
Converting the Ontologies to OWL

• For each Data Property
• ltowlDatatypeProperty rdfID"length"gt
• ltrdfsdomain rdfresource"Road"/gt
• ltrdfsrange rdfresource"xsdpositiveInteger"/gt
• lt/owlDatatypePropertygt
• For each Object Property
• ltowlObjectProperty rdfID"spannedBy"gt
• ltrdfsdomain rdfresource"Bridge"/gt
• ltrdfsrange rdfresource"Lake"/gt
• lt/owlObjectPropertygt

19
Converting the Ontology Mappings to OWL

• Create a owlOntology element with the imports
  for Source and Target ontology.
• ltowlOntology rdfabout""gt
• ltrdfslabelgt Integration of OGC and Ordnance
  Survey
• ltowlimports rdfresource"S229.owl" /gt
• ltowlimports rdfresource"S9.owl" /gt
• lt/owlOntologygt
• For each class in the source ontology get it
  mappings
• If the mapping is of the type equivalent then
• ltowlClass rdfabout"RoadNode"gt
• ltowlequivalentClass rdfresource"RoadPoint"/gt
• lt/owlClassgt

20
Converting the Ontology Mappings to OWL

• 1. If the mapping is of the type SubClassOf then
• ltowlClass rdfabout"RoadNode"gt
• ltrdfssubClassOf rdfresource"RoadPoint"/gt
• lt/owlClassgt
• 2. If the mapping is of type SuperClassOf then
• ltowlClass rdfabout"RoadPoint"gt
• ltrdfssubClassOf rdfresource"RoadNode"/gt
• lt/owlClassgt

21
Converting the Ontology Mappings to OWL

• 1. For each Data Property in the source ontology
• ltowlDatatypeProperty rdfabout""gt
• ltowlequivalentProperty rdfresource"RoadPoint"/
  gt
• lt/owlDatatypePropertygt
• 2. For each Object Property in source ontology
• ltowlObjectProperty rdfabout""gt
• ltowlequivalentProperty rdfresource"RoadPoint"/
  gt
• lt/owlObjectPropertygt

22
Wrapper Generation

• Ontology Mappings can be seen as abstract
  specification of relationships between ontology
  entities
• Wrapper gives the rules for converting the data
  according to one schema into that of according to
  another schema.
• Data from different sources will be in GML form,
  which is XML based.
• So, we generate XQuery transformations for
  achieving this.

23
Source Ontology
Target Ontology
Ontology
Wrapper
Mapper
24
Wrapper Generation
Onto1.feature1 Onto2.feature1
Onto1.feature1 Onto2.feature2
. . . . . . .
Rules written in xQuery format
Rules generated by semantic matching
(semi-automatic)
Onto1.featureM Onto2.featureN
25
Attribute - Simple Equivalence
XQuery Template
26
Attribute - Transformation
XQuery Template
27
Class - Simple Equivalence
XQuery Template
28
Class - Subclass
XQuery Template
29
Class - Superclass
XQuery Template
30
Wrapper Generation

• Start with Source node call the
  RecursiveXQueryGen
• If Source
• 1 Get all First level layers
• 2 Get all direct Classes of the source
• 3 Start the XQuery template for Source
• 4 Take each layer collected and recurse
• 5 Take each class collected and recurse
• 6 End template for Source

31
Wrapper Generation

• If Layer
• 1. Get all its Sub layers
• 2. Get all its Classes
• 3. Check for its Mappings
• 4. Start the xQuery template for each layer
  accordingly
• 5. Take each layer collected and recurse
• 6. Take each class collected and recurse
• 7. End template for Layer

32
Wrapper Generation

• If Class
• 1. Get all its Sub Classes
• 2. Get all its Attributes
• 3. Check for its Mappings
• 4. Start xQuery template for each class
  accordingly
• 5. Take each attribute collected and recurse
• 6. End template for Class
• 7. Take each class collected and recurse

33
Wrapper Generation

• If Attribute
• 1. Get the Mapping type
• 2. Process the User Transformation
• 3. Generate the XQuery accordingly

34
Data Extraction
Wrapper
Data1, Data2
(Changes the structure of data)
Source Ontology Source Data
Data1, Data2
Target Ontology Source Data
35
Data Extraction

• Data Extraction involves getting the actual data
  from the
• data sources and then applying the
  transformation.
• We provide an interface for querying the data at
  feature level.
• For the features requested, their mappings are
  identified and then WFS Requests are generated
  for different data sources to get the GML data.
• Transformations are applied on the data got from
  different sources and the merged data is given to
  the user.

36
Data Request and its Transformation

• Data request will consists of
• a) list of features
• b) Coordinates of bounding box
• Identify the data sources with which the data
  consumers ontology is mapped.
• For each data source with which there exists a
  ontology
• mapping,
• Check mappings for the features in the data
  request.
• Add all the features in mappings for that
  particular source.

37
Experiments

• Experiments using ontologies of Ordnance Survey
  and OGC for the transportation domain
• Lexical and Propagated similarity values are
  taken and compared with human generated values
• For human generated similarity values three
  Scores for every pair of classes
• Based on English meaning of the words.
• Reflexive of Superclass similarity
• Reflexive of Subclass and attribute similarity

38
Analysis

• False Positives (the error of rejecting
  something that should have been accepted e.g.,
  such as finding an innocent person guilty)
• 12.3 at t threshold of 0.50
• 36.9 at a threshold of 0.40.
• Mainly seen in the cases where parts of the
  target class name existed as a part of the
  source class name.
• False Negatives (the error of not rejecting
  something that should have been rejected e.g.,
  such as finding a guilty person innocent)
• 4 percent at a threshold of 0.30
• (25) in the top 20 class matches based on human
  generated similarity values.

39
Analysis
40
Conclusion

• Semantic Heterogeneity is the key challenges for
  GIS Interoperability.
• Lexical similarities of schema element labels and
  descriptions can help in ontology mapping.
• Similarity propagation based on heuristics allows
• integration of implicit semantics of the
  ontology structure and hence improves the
  mapping.
• Mapping Validation, Wrapper Generation, Data
  Extraction components further help in having a
  End-to-End framework for GIS Interoperability

41

• Thank You