Affinity-based Schema Matching

1
Affinity-based Schema Matching

• Silvana Castano
• Università di Milano

D2I Modena, 27 aprile 2001
2
Affinity relationships

• Schema-level matching (no instances)
• Find mappings between schema elements that
  correspond semantically to each other
• Affinity mappings have an associated calculated
  degree of similarity
• ARTEMIS is the schema matching component of the
  MOMIS system which evaluates schema element
  (e.g., ODLI3 classes) affinity based on the
  following comparison features
• class names
• class attributes (name and domain)
• class references (name and domain)

3
Affinity Coefficients

• Name Affinity ?0,1
• Names are compared by exploiting knowledge
  provided by a reference ontology O (e.g., the
  Common Thesaurus)
• Ontology O is organized according to given
  strengthened terminological relationships (e.g.,
  synonymy, hypernymy).
• Given two names n and n an affinity function
  A(n,n)?0,1 returns the strength of the path
  between them in O they have affinity iff
  A(n,n) is greater than a

4
Name Affinity
Thesaurus Definition
Determining the path with highest strength
School_Member
Computing Affinity coefficient
NA(School_Member, Professor) 0.8 0.8 0.64
5
Affinity Coefficients

• Structural Affinity ?0,1
• Domains are compared based on compatibility
  relationships (e.g., validated attributes in the
  Common Thesaurus)
• The coefficient is proportional to the number of
  matching attributes and referenced classes
• Global Affinity ?0,1
• Comprehensive affinity value, weighted sum of the
  two previous coefficients

6
Structural Affinity(Interactive validation)
ODLI3 Classes
Semantic Correspondences
Validity Check
SA(School_Member, Professor) 0.25
7
Affinity-based Clustering

• Hierachical clustering techniques are employed to
  identify all schema elements candidates to
  integration based on evaluated affinities.
• Clusters of candidates are selected interactively
  based on affinity thresholds
• Clusters selection occurs in a way that schema
  elements inside a cluster are characterized by
  high values of affinity with other elements of
  the cluster and by lower levels of affinity with
  elements outside.

8
Affinity Tree Candidate Clusters
Threshold choice (es. 0,6)
Candidate cluster choice
9
Ongoing work and open issues

• Application of the affinity/clustering schema
  matching techniques to integration of XML
  datasources has been started by Milano and first
  results have been published IWASS00,Retis01,DS-90
  1. Ongoing work is on XML data reconciliation by
  providing a further ontological layer on top of
  integrated representations.
• Affinity and clustering are actually performed
  based on intensional inter-schema properties
• Extension of affinity and clustering techniques
  to consider also extensional inter-schema
  properties will be performed in collaboration
  with Modena and Reggio-Emilia.