Databases Reference
In-Depth Information
(a) Graph for A-Box
A
1
(b) Graph for A-Box
A
2
Fig. 3.
Examples of two different A-Boxes representing triples in RDF. Dashed arrows
indicate object properties that occur in both A-Boxes.
constraints from the ontologies should be taken into account in the matching
process. In particular, objects should not be maps on each other if they have
incompatible types. In the example this means that assuming the underlying
ontology contains a statement
student⊥pupil
declaring the classes 'student' and
'pupil' as disjoint, the objects
a
3
and
b
3
should not be mapped on each other,
despite the high a priori similarity.
4 Statistical Relational Learning
Data integration for heterogeneous knowledge bases typically involves both
purely logical and uncertain data. For instance, the description logic axioms
of the ontologies are known to be true and, therefore, should be modeled as
logical rules - the alignment system should not alter the logical structure of the
input ontologies. Conversely, matching systems usually rely on degrees of confi-
dence that have been derived through the application of lexical similarity, data
