![](bg2.png)
2.2 Similarity Measures
Ontology is described by structure of concepts
which the relation of subsumption (subClassOf) is
the primary relationship. This structure defines the
semantics of these concepts. The measures that
exploit this structure are called semantic measures of
concepts. Thus, Semantic measures can be used to
assess a link between two concepts of the same
ontology by exploiting their relationship.
Blanchard et al (Blanchard et al, 2008a)
classified semantic similarity measure in to three
types: measures that focus in the characteristic of
ontology’s entities, semantic relationship measures
and informational content measures.
For the first, the similarity between two concepts
is defined based on both common and
different
characteristics of those two concepts (Dice, 1945).
For the second, metric are proposed to measure
conceptual distance between two concepts of the
same ontology which is computed based on the
number of edges separating these two concepts
(Rada, 1989) or based on mscs(Ci;Cj ) which refers
to the most specific subsume (the lowest common
ancestor in the tree) of both concepts Ci and Cj) (Wu
and Palmer, 1994), or else improving measurement
accuracy by considering other semantic links in
addition to subsumption (Ganesan et al ,2003)
( Maguitman et al, 2005).
The third type, based on informational content,
distinguishes between two categories of measures.
The first one is based on textual corpus which
associate a probability P with concepts in a “is-a”
hierarchy to denote the likelihood of encountering an
instance of a concept c in a textual corpus.and others
using ontology structure (Resnik, 1999).
For The second category, (Blanchard et al,
2008b) present new method for computing the
information content of concept by considering only
the taxonomic structure of the ontology. Otherwise,
(Blanchard et al, 2008b) proposes four hypothesis of
instance distributions which used to compute the
informational content of a concept.
The same authors (Blanchar et al, 2008b)
propose a new measure PSS “the Proportion of
Shared Specificity” which takes into account the
density of links in the graph between two concepts.
This measure is based on one of the hypothesis
described above and called Ps. This hypothesis
implies an uniform distribution among the set of
sons of each concept, the informational content of a
concept depends on the number of sibling of the
subsuming concepts.
The enrichment approach based on stability
assessment we that we are going to propose can
apply various similarity measures in particular the
PSS measure.
3 STABILITY EVALUATION
There many approaches for ontology assessment, a
survey is described in (Brank et al, 2005). We think
that the most useful approach of ontology quality
evaluation is the one based on the use of the
ontology in real world application. The user, who
interacts with ontology based system, is interested in
the response to their request queries. So, we look for
the stability of the results regarding ontology
evolution with evaluating the semantic and structural
change between initial ontology and its enrichment.
It is evaluated based on semantic relation between
concepts of ontology. Thus, when the stability is
reached, the ontology will still with the same
semantic structure. This will lead to the same
response to user queries through enrichment.
The ontology stability according to the enrichment is
considered as semantic difference between initial
ontology and enriched one. The semantic difference
can be computed relatively to similarity between
concepts which evaluate its cohesion.
The stability is
computed using the average of the similarities
between the concepts of different ontologies (O
1
as
initial ontology and O
2
is the enrichment of O
1
).
∑∑
==
−
=
n
i
n
j
O
j
O
i
O
j
O
i
n
ccsimccsim
OOStability
11
2
21
),(),(
),(
2211
(1)
where n is the cardinality or the number of concepts
contained in O
1
and O
2
is the enrichment result of
O
1
)(
21
OO ⊂
.
1
O
i
C
represents the concept
i
C
in
ontology O
1
and Sim is the semantic similarity
measure between two concepts. We choose the
information content PSS (Proportion of shared
specificity) as similarity measure (Blanchar et al,
2008b). If the function of stability tends to 0, the
ontology evolution will be considered to be perfect
and don’t affect the stability of the ontology.
4 ENRICHMENT APPROACH
We propose a new approach for adding new
concepts to ontology. It should consider the stability
and semantic relation to get the right way for
enrichment. Indeed, adding new concepts must be
with minimizing the affect on the structure and the
semantic of ontology. It is made by the following
AUTOMATICAPPROACHFORONTOLOGYEVOLUTIONBASEDONSTABILITYEVALUATION
453