Determining Semantic Similarity among Entity Classes from Different Ontologies

1
Determining Semantic Similarity among Entity
Classes from Different Ontologies

• M.Andrea Rodriguez
• Max J Egenhofer
• Presented By
• Chintan Parekh

2
MOTIVATION ?

• Data becoming increasing complex and
  heterogeneous.
• We have complex knowledge management task.
• Solution
• Need to understand the underlying semantics of
  data.
• Use of Ontologies (knowledge based systems) and
  semantic similarity functions.
• Improve the retrieval and integration of
  information.
• Focus on ontology which define entity class,
  relations among classes and distinguishing
  features among classes.

3
Introduction Semantic Similarity

• Traditional Approach
• Compute semantic distance between
  definition within a single ontology.
• Current Approach
• Relax the requirement of a single ontology and
  account for the differences in level of
  explicitness and formalization of the different
  ontology specifications.
• Determine Similar Entity classes (objects in
  real life) with the use of synonym sets, semantic
  neighborhood and distinguishing features.

4
Type of ontology Approach Used

• Terminological Ontology
• No use of Axioms. Use relations sub-type,
  super-type, is-a ,part-whole.
• Axiomatic Ontology
• Kind of terminological Ontology. Use of Axioms
  and definition stated in logic.
• Approach
• Modeling similarity based on matching
  process that uses information from different
  ontology specifications.
• Eg.Synonym sets, semantic relations of
  entity classes.
• Used to determine which entity classes are
  most similar and can be integrated across
  different ontologies

5
Current Approaches

• Map the local terms of distinct ontology onto a
  single ontology.
• Semantic similarity determined as function of
  path distance between terms in hierarchical
  structure as in single ontology.
• 2. Create a shared Ontology by integrating
  existing ones.
• ONIONS Methodology for Ontology Analysis and
  Integration.
• 3. Use of semantic Inter-relations
• OBSERVER Ontology based system that is enhanced
  with relationships for vocabulary heterogeneity
  resolution.

6
Current Approaches (Contd..)

• Uses Synonymy and hyponymy
• Bergamashi Et Al Uses Synonymy and hyponymy
  terminology for ontology integration.
• Measure for comparing concepts (after ontology
  integration)
• 1. Filter measures based on path distance
• 2. Matching measures based on Graph (1-1
  Correspondences)
• 3. Probabilistic Measures.

7
Entity Class Representation

• Set of Synonym words that denote an entity class
• Polysemy Same word having more than one meaning
  
• Synonymy Different words having same meaning
• Set of interrelations among entity classes
• 1.Hyponymy IS-A relationship
• This is transitive and asymmetric
  (Inheritance from super ordinate)
• 2. Meronymy PART-WHOLE relationship
• Transitive property holds for some but
  not all

8
Entity Class Representation (Contd)

• 3. Set of Distinguishing feature that
  characterize entity class
• 1. Sometimes the IS-A relationship is not
  enough for the distinguishing
• entity class
• 2. Attributes used for describing distinguishing
  features of
• entity class
• 3. Features classified as
• a) Functions Functionality
• b) Parts Structural Elements
• c) Attributes Additional Characteristic
  of entity class

9
Example of Entity Class
10
Components of Similarity Assessment

• 1.First assessment deals with similarity of
  synonym sets.
• a) We know these are grouping of
  semantically equivalent.
• b) Cross Ontology agreement in use of words
  and detect equivalent words that likely to refer
  to same entity class
• E.g.. Clinic and Hospital have high
  semantic similarity
• 2.Second Assessment is the distinguishing
  features of entity class.
• E.g.. Stadium and Sports arena Place where
  people play sports (Similar concept)
• 3.Third Assessment involves semantic relations
• a) The relations become comparison between
  semantic neighbourhood of entity classes

11
Components of Similarity Assessment (Contd..)

• b) E.g.. Hospital and House are related to
  super class Building (so semantically similar)
• Use of weights sum of each specification
  component

12
Example of Semantic Neighbourhood
13
Matching Based Approach to Model Similarity

• Set Theory Model
• Produce a Similarity value which is a result of
  common as well as different characteristics of
  objects
• S(a,b) AnB/(AnB a (a,b)A/B1-
  a(a,b))B/A)
• where
• (A/B) gt Difference
• (AnB) gt Intersection
• A,B gtDescription sets of a and b (Set of Synonym
  Set etc)
• a gt Relative Importance of noncommon
  characteristic
• 0 ltalt1
• gt Cardinality

14
Examples of Similarity Assessment

• Use of two Ontologies Wordnet and SDTS
• Word Matching
• Gives a value of S1 if same word used in
  both ontologies
• Gives a value of S0.58 if similar words used
  in these ontologies
• Feature Matching
• Matching based on parts ,functions and
  attributes
• Each given equal weight (Default)
• When no classification then a global feature
  matching used.
• Semantic Neighbourhood Matching
• Compare Entity classes based on word or
  feature matching.
• Depends on the radius of semantic
  neighborhood.

15
Cross Ontology Evaluation

• No work done on matching to correlate
  computational method and peoples judgments
• This study about experimentally creating a
  correlation among different ontologies
• Experiments Used for similarity Evaluation
• 1)Similarity evaluation across independent
  ontologies
• 2)Model which creates automatically association
  across ontologies
• 3)Use available ontology i.e. WordNet and SDTS
• 4)Use human subject testing to be used as
  benchmark

16
Cross Ontology Evaluation (Contd..)

• A new ontology developed for purpose of
  experiments
• Ontology called WS by the combination of WordNet
  and SDTS
• Features of WS
• 1) Complete definition of entity class.
• 2) Part Whole and is-a relationship
  included.
• Goals of Experiments
• 1) Searching for similarity among entity
  classes.
• 2) Ranking of similarity.

17
Experiment I

• To evaluate result Use of recall and precision.
• Recall gtFraction of Similar Entity classes
  detected
• Precision gt Fraction of Similar Entity
  classes detected that are actually similar
• Considered Spatial entity class present on
  campus map
• RESULTS
• Compared Entity classes across ontologies using
  different weights.
• Result are highly sensitive to components of
  entity class representation.
• Similarity functions should share those component
  that are common to all ontologies.

18
Result Analysis
19
Experiment II

• Cross ontology evaluation transformed into a rank
  of similarity.
• Compare Entity class in a ontology with a reduced
  set of entity class in another ontology.
• Ontology Comparison Done
• 1)WordNet WS
• 2)SDTS WS
• 3)WS-WS
• Human Subjects rank the similarity among entity
  class based on definition in WS.
• RESULT
• The Performance of model depends on
  compatibility of ontologies

20
RESULTS
21
Conclusion Future Work

• The similarity model is systematic way of
  detecting similar entity classes across
  ontologies.
• Similarity model first step in ontology
  integration.
• Future Work
• 1)Parts are entity classes could be compared.
• 2)Attributes in terms of domains could be
  compared.

22

• Thank You !!!!!!!