Semantic Similarity based on Weighted Ontology - Geographical Information Systems: Trends and Technologies - page 315

Table 3. contd.

Pairs of classes HJ Lin Dice MDSM v MDSM c tSim GSim v GSim c

(Department, Province) 0.5700 0.5393 0.2857 0.1620 0.1230 0.2500 0.3871 0.4114

(Department, Region) 0.7100 0.6212 0.2857 0.1543 0.1171 0.2500 0.4259 0.4571

(Department, Country)

0.8200 0.6724 0.2500 0.1350 0.1025 0.2500 0.4502 0.4857

(Department, State) 0.7300 0.6718 0.2857 0.1543 0.1171 0.2500 0.4500 0.4854

(Department, County) 0.7200 0.5993 0.5714 0.3240 0.2460 0.5000 0.5471 0.5554

Correlation

1.0000 0.7127 0.2972 0.3774 0.3419 0.3318 0.7364 0.8037

We observe that the highest correlation of similarity measure with
HJ is achieved by GSim c (0.8037), and the correlation of GSim v (0.7364)
with HJ is slightly higher than the correlation of Lin (0.7127). We also
observe a signifi cant difference between similarity scores obtained by the
hierarchical-based approach of Lin and the feature-based approaches of
Dice and MDSM . The scores obtained by the combined approach GSim,
in most cases are greater than the values computed according to Dice and
MDSM , and they are less than the measures calculated according to the
information content approach by Lin . The reason is twofold. Firstly, similar
to Dice and MDSM , the GSim approach considers the structure of classes,
and the heterogeneity of the attributes, in some cases, signifi cantly impacts
on the similarity values that are considerably less than the ones obtained
according to Lin . Secondly, the GSim method captures the informativeness
of geographic classes organized as a hierarchy, while Dice and MDSM are
mainly based on the common and non-common characteristics of classes.
In Table 3, some scores by Dice and MDSM are equal to zero, indicating
that the pairs of considered classes have no common features and they are
completely distinct. Similar results we obtain by tSim component of GSim ,
which is addressed to capture the similarity of classes' structures.

Analysis of parameters

In this section, we analyze the variability and commonality parameters in
the combined GSim method. These parameters change depending on the
variation of the number of occurrences of attributes and parts in formulas
(5), and (6). To this end, we perform two experiments. In the fi rst experiment,
we change attributes occurrences, and fi x the occurrences of parts, while
in the second one we change parts' occurrences and fi x the occurrences of
attributes.
We start with the fi rst experiment, for which we fi x the number of
occurrences of parts, say 19, and change occurrences of attributes. Intuitively,
as the number of occurrences of attributes increases, consequently, the
number of common attributes between classes increases as well. This