SEKE 2012 Proceedings - Knowledge Systems Institute

if (type(c)=”class”) then
add c to groups cls
else if (type(c)=”property”) then
add c to groups props
end if
end if
end for
classify (groups cls
)
classify (groups props
)
The generic classif ication phase classify(groups in ) is the
following:
//classify(group in
)
if(size(group in
)>1) then
remove concept c r
from group in
for ( c group ) do in
if ( cr
Oi
_ and _ c O jwith
_ i j )then
if(type(c r
) = “class” and type(c)=”class”)then
index = calculateIndexSimilarityClasses(c r
, c)
else
if(type(c r
) = “property” and type(c)=”property”) then
index = calculateIndexSimilarityProperties(c r
, c)
end if
if(index>threshold) then
add mapping found between c r
and c
end if
end if
end for
classify(group in
)
end if
According to the proposed approach, the mapping process
is a cl assification problem: it classifies the similarities
among the classes, the properties and the relationships
and then creates a new on tology that is a common layer
representing a shared view of the various ontologies. As
previously said, various approaches are in literature ([3]-
[11]) in order to find the semantically similar components
in the ontologies. The a dopted functions are the
following:
Editing Distance (ED): This function is so defined:
min( x , y ) ed(
x,
y)
sim ed
( x,
y)
1
max(0,
) [0,1]
min( x , y )
It aims to calculate the li kelihood among words that
labelling concepts in t he ontology. In particular it
compares the syntactical structure of the words and counts
the characters that are in the same position in the words x
and y by the use of the “ed” function. The value 1 means
that the two words are similar.
Trigram Function (TF): This function aims to measure
the number of similar trigrams that are in the words that
label the concepts in the ontologies.
1
TF( x,
y)
[0,1]
1
tri(
x)
tri(
y)
2 * tri(
x)
tri(
y)
The function tri(x) gives the set of trigrams that are in the
in the word x. The value 1 means that the two words are
similar.
Semantic similarity index (SS): This index is so defined
1
SS(
w1 , w2
)
[0,1]
sim ( w , w )
where
jc
1
2
sim jc (w 1 , w 2 ) = 2* log P[LSuper(c 1 , c 2 )]-[log P(c 1 )+ log P(c 2 )]
This index aims to compare from a semantic point of view
two words measuring their distance in t he taxonomy
defined in Wordnet [16]. The value 1 means that the two
words are similar.
Granularity (GR): This i ndex measures the mutual
position of the words representing the concepts of the
ontology in the W ordNet taxonomy. This index is so
defined:
min[ dens(
c1
)* path(
c1,
p),
dens(
c2
)* path(
c2,
p)]
GR(
c , c2
)
max[ dens(
c )* path(
c , p),
dens(
c )* path(
c , p)]
1
1
1
2
2
[0,1]
where dens(c) is the function representing the density of
the concept c. This function is defined as E(c)/E where E
is the ratio between the number’s arc of the concept and
the numbers of its parents while E(c) is the number of the
sibling of the concept c. Th e function path(c 1 , p) is the
shortest path from c 1 to p that is the first parent common
to c 2 .
Attribute Index: This index aims to measure the
numbers of similar attributes between two nodes. In
particular, it is so defined:
sim att
X Y
X
X Y
(
x,
y)*
(1 (
x,
y))*
Y
Y
X
[0,1]
with [0,1]
is a parameter and X and Y are the number
of attributes related to the two compared nodes. If this
function gives value 1 it means that the words X and Y
are similar.
Synonym Index (SI): This index aims to verify if in
Wordnet there are synonyms of the word related to th e
concept in an ontology that label a concept in another
ontology. This index can assume value 0 (no synonym) or
1 (synonym).
Derived Index (DE): This index aims to find in WordNet
an adjective, representing a node of ontology, derived
from the label of a concept that is in the other ontology.
This index can assume value 0 (not derived) or 1
(derived).
Property Similarity Index (ISP): This index has the aim
to verify the equality between the nodes evaluating their
181