Semantic interoperability through analysis of model extension

1
Semantic interoperability through analysis of
model extension

• Xiaomeng Su
• 18-10-2002 Trondheim

2
Semantic interoperability in heterogeneous systems

• Achieving semantic interoperability typically
  involves resolving the semantics of the data in
  the separate sources.
• A conceptual schema (model, ontology) usually
  carries the semantics of data.
• Reconciling data semantics is normally based on
  first identifying similar entities in various
  model and then reconciling them.
• Semantic reconciliation was traditionally studied
  in database system.
• Semantic reconciliation in semi structured source
  also attracts much research recently.

3
An example

• Product catalog integration in e-commerce
• Two classification systems
• UNSPSC (United Nation Standard Products and
  Services Code System)
• Ecl_at_ss
• ECCMA (Electronic Commerce Code Management
  Association)

4
UNSPSC
5
Ecl_at_ss
6
Problem statement

• Assumptions
• An ontology is a set of elements connected by
  some structure. Among the structures, we single
  out hierarchical IS-A-relation and all the others
  we call them related relations, which is merely
  an indication of relatedness. A classification
  hierarchy is a typical example of ontology
  organized only by hierarchical IS-A-relation.
• There exist different ontology representational
  languages Su02, we assume that it is possible
  to translate between different formats. In
  practice, a particular representation must be
  chosen for the input ontologies. Our approach is
  based on Referent Modelling Language (RML)
  Soelvberg98, which is an ER-like language with
  strong abstraction mechanism and sound formal
  basis.
• Research statement.
• Given two ontologies A and B, mapping one
  ontology with another means that for each concept
  (node) in ontology A, try to find a corresponding
  concept (node), which has same or similar
  semantics, in ontology B and vice verse. To be
  more exact, we need to
• a)define the semantic relationships that can
  exist between two related concepts.
• b)develop algorithm, which can discover concepts
  that have similar semantic meaning.

7
A candidate solution

• Believe.
• the semantic meaning of a category (node) is best
  described by the documents that have already been
  categorized under that category
• the process of mapping ontologies can be
  supported by analysing the extension of concepts
  to derive corresponding intentional descriptions.
• Enriching the semantic meaning of each node by
  looking at the instances (documents) it
  possesses.
• Represent the node as a generalization of the
  information its instances provide.
• Calculating similarity based on the
  generalizations.

8
An architecture for ontology mapping through
extensional analysis
9
System functional view

• Text categorization (CnS prototype)
• Feature vector construction.
• Pre-processing
• Document representation
• Feature selection
• Concept node vector construction
• Similarity calculation
• Mapping assertion generation

10
Vector similarity simple introduction

• Given a query
• and a document collection
• The similarities are

11
Example normalized similarities

• Given a query
• and a document collection
• The similarities are

q ? d q ? (d/dw)
d1 d2 d3 d4
2 2/2 1 2
2/4 0.5 0 0/2 0
3 3/4 0.75
12
Cosine similarity

• Measuring the angle between vectors

Q
?
D2
cos (?)
Angle 90 -gt cos (?) 0 no similarity Angle 0
-gt cos (?) 1 Max similarity (equal)
13
Cosine similarity

• Let a query vector q and a document vector d,
  both of lenght n, be given.
• The cosine similarity is defined as

14
Term weights

• Up to now, we only considered binary term
  weights
• 1 term occurs in documents
• 0 term does not occur in document
• Two shortcomings
• Does not reflect frequency of terms
• All terms are equally important (e.g. president
  vs. the)
• Improvement
• Store the frequency in the vector (e.g. 4)
  instead of 1
• tf-score
• Meaningful terms occur only in a few documents,
  they are good discriminators - distinguish those
  documents from the rest
• idf-score

15
Feature vector construction

• Pre-processing
• remove (html) tags, remove stop word
• Represent document as vectors.
• Remove non-informative feature (for dimension
  reduction.)
• Calculate feature vector for each node.
• Leaf nodeas an average vector of its possessing
  documents vectors
• Non-leaf node taking into consideration
  contributions from the documents that have been
  assigned to it, its direct sub nodes and the
  nodes to which node K hold a related relation

16
Similarity calculation

• Using cosine measure
• using just one approach is unlikely to achieve as
  many good mapping candidates as one that combines
  several approaches. Therefore our approach should
  factor into other information as well
• Linguistic information
• User specified information.

17
Mapping assertion
18
(No Transcript)
19
Way of working

• 1) Survey of ontology mapping methods and
  analysis of the ontology mapping process.
• 2) Survey of applicable parts of text
  categorization and information retrieval.
• 3) Development of an ontology mapping algorithm
  based on text categorization techniques.
• 4) Application of 3) in a case study
• 5) Analysis of empirical observations from 4) and
  evaluating its usage.