SEMANTIC CLUSTERING BASED ON ONTOLOGIES
An Application to the Study of Visitors in a Natural Reserve
Montserrat Batet, Aïda Valls
Department of Computer Science and Mathematics, Universitat Rovira i Virgili, Av. Països Catalans 26, Tarragona, Spain
Karina Gibert
Department of Statistics and Operations Research, Univ. Politècnica de Catalunya, C. Jordi Girona 1-3, Barcelona, Spain
Keywords: Ontologies, Numerical and categorical data, Semantic similarity, Clustering, Tourism analysis.
Abstract: The development of large ontologies for general and specific domains provides new tools to improve the
quality of data mining techniques such as clustering. In this paper we explain how to improve clustering
results by exploiting the semantics of categorical data by means of ontologies and how this semantics can be
included into a hierarchical clustering method. We want to prove that when the conceptual meaning of the
values is taken into account, it is possible to find a better interpretation of the clusters. This is demonstrated
with the analysis of real data collected from visitors to of a Natural Reserve. The results of our methodology
are compared with the ones obtained with a classical multivariate analysis done in the same database.
1 INTRODUCTION
Nowadays, the extensive use of information and
communication technologies provides access to a
large amount of data (e.g. Wikipedia, electronic
questionnaires). Usually, these resources provide
textual data that may be semantically interpreted
(e.g. a questionnaire can ask about the “Main hobby”
of the responder, whose answer could be dancing or
trekking). Classically, in clustering these terms were
managed as categorical features (nominal or
ordinal), whose values are expressed with labels in a
predefined set of terms. Categorical values are
compared at a syntactic or ordinal level, but without
a semantic analysis. Taking into account the
conceptual meaning of those terms, more accurate
estimations of their degree of similarity (e.g.
trekking is more similar to jogging than dancing)
should be obtained, improving the quality and
interpretability of the clusters obtained.
On one hand, this paper presents a hierarchical
clustering method that can deal with numerical,
categorical and with variables for which semantic
information is available, named semantic features.
In the core of the clustering algorithm, comparisons
between objects are done with a compatibility
measure which takes into account the different types
of features and permits to make a homogeneous
treatment. To provide the semantic knowledge,
ontologies will be considered and use used to
compare semantic features.
On the other hand, we test this clustering method
with a dataset obtained from visitors of the Ebre
Delta Natural Park. The goal is to find a
characterization of the types of visitors in terms of
their tourist and trip profiles. The results of our
methodology are compared with the ones obtained
with a statistical analysis made in with the same
database. Results show that the clusters that consider
the semantics of the terms give a more valuable
classification of the visitors.
The paper is organized as follows: in Section 2,
the clustering method for numerical, categorical and
semantic features is presented. In Section 3, a
dataset is introduced and an analysis of the results of
a previous study is done. In section 4, the same data
set is tested using the proposed method; the results
are studied. The last section gives the conclusions.
2 SEMANTIC CLUSTERING
To include the semantic aspect into the clustering
process, semantic features are introduced.
283
Batet M., Valls A. and Gibert K..
SEMANTIC CLUSTERING BASED ON ONTOLOGIES - An Application to the Study of Visitors in a Natural Reserve.
DOI: 10.5220/0003165602830289
In Proceedings of the 3rd International Conference on Agents and Artificial Intelligence (ICAART-2011), pages 283-289
ISBN: 978-989-8425-40-9
Copyright
c
 2011 SCITEPRESS (Science and Technology Publications, Lda.)

 A feature X
k
is a semantic feature if: X
k
takes
linguistic values.
 Linguistic values or terms appearing in X
k
can be
semantically compared exploiting some background,
like ontologies.
Since the values of semantic features became
concepts (i.e. they correspond to labels of concepts
in the reference ontology (Studer, Benjamins, and
Fensel, 1998)) rather than simple modalities, it is
possible to perform comparisons between values
using a semantic similarity function.
2.1 Measuring Semantic Similarity
The definition of a measure of distance/similarity
between the values of a pair of semantic features is
essential for comparing objects in a semantic
clustering approach.
This similarity is quantified by determining how
concepts are alike based on semantic evidences
observed in some knowledge source (e.g. a ontology
or a corpus). According to the knowledge exploited
in order to estimate the similarity between terms
these functions can be classified in different
families.
In some methods, taxonomies and, more
generally, ontologies (Studer et al., 1998) are
considered as a graph model in which semantic
relations are modelled as is-a links between
concepts. Then, the similarity is usually a function
of the minimum number of is-a links (i.e. minimum
path) between concepts (Rada, Mili, Bichnell, and
Blettner, 1989). Similarity can also be a function of
other features such as the depth of the concepts in
the taxonomy (Leacock and Chodorow, 1998). The
main advantage of these Taxonomy-based measures
is that they only rely in a ontology for assessing the
similarity. However, they are affected by their
dependency on the degree of completeness,
homogeneity and coverage of the ontology.
Other approaches consider not only the ontology
but also the distribution of the compared terms in a
corpus. These approaches rely on the Information
Content (IC) of concepts (the inverse to its
probability of occurrence in a corpus). Similarity is
usually estimated as the IC of the first common
ancestor of the compared concepts (Resnik, 1995)
(Lin, 1998). In general, IC based measures provide
better results than Taxonomy-based measures as they
exploit a great amount of knowledge. However, they
are affected by data sparseness if there is not enough
available data to estimate information distribution.
In previous works (Batet, Sanchez, Valls, and
Gibert, 2010) we have done an extensive study of
the characteristics, performance, advantages and
drawbacks of several semantic similarity measures.
As a result of this study we propose the use the
similarity measure presented in (Batet et al., 2010).
This measure is based on the exploitation of all the
taxonomical knowledge in a ontology (i.e. the full
set of ancestors of a pair of compared concepts). In
fact, the measure is based on the ratio between the
number of different is-a ancestors of terms c
1
and c
2
in a reference ontology and the total number of
ancestors of the terms, where A(c
1
) and A(c
2
) are the
is-a ancestors of c
1
and c
2
in the ontology including
themselves, respectively.
)Α(c)Α(c
cΑcΑcΑcΑ
cc
SCD
d
21
2121
2,1
)()()()(
log)(



(1)
On the contrary to previous approaches, where only
a partial view of the modelled knowledge of the
ontology is considered (i.e. the minimum path
between concepts), this measure considers the
relationships given by multiple inheritance of the
concepts. This approach has been proven to provide
a more accurate estimation of the similarity than
classical Taxonomy based and IC based measures
(Batet et al., 2010). In addition, as this measure only
rely on the ontology, it has a low computationally
cost. Therefore, in the semantic clustering proposal
that we present, this measure is applied for
evaluating semantic features.
2.2 Object Comparison
with Numerical, Categorical
and Semantic Features
Typically, an object is represented by a
multidimensional vector where each dimension
represents a feature or variable. Here, the case in
which features can indistinctly be either numerical
or categorical or semantic is treated. A compatibility
measure for comparisons is introduced.
Metrics for mixed numerical and categorical
values can be found in the literature (Ahmad and
Dey, 2007; Gibert and Cortés, 1997). However,
authors are not aware of references including also
semantic features. In this sense, our proposal
generalizes Gibert’s mixed metrics (Gibert and
Cortés, 1997) including semantic features.
Data is represented as a matrix, where objects I
= {1,…,n} are in the rows, while the K features
X
1
…X
K
are in the columns
.
Thus, each cell (x
ik
)
contains the value taken by object i for feature X
k
.
The distance between a pair of objects i and i', is
calculated as the combination of applying a specific
ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence
284

distance for each type of feature X
k
. This distance is
defined in eq. 2:

1 ,1,0),,(with
)',()',()',()',(
3
222
),,(
2






iidiidiidiid
SQ
(2)
where ζ = {k : X
k
is a numerical feature, k =1:K}, Q
= {k : X
k
is a categorical feature, k =1:K}, and S = {k
: X
k
is a semantic feature, k =1:K}. In our proposal
d
ζ
2
(i,i’) is the normalized Euclidean distance for
numerical features, d
Q
2
(i,i’) is the χ
2
metrics for
categorical values and d
S
2
(i,i') is the similarity
measure introduced in the previous section.
In (2) each component has an associated weight.
The weighting constants (α,β,γ) are taken as
functions of the features’ characteristics. In
particular, they depend on the range of distances of
each type of feature and how many variables refer.
We have (α,β,γ)
 [0,1]
3
with α+β+γ =1, being
n
ζ
=card(ζ), n
Q
=card(Q), n
S
=card(S) and
max*
2

d
,
max*
2
Q
d
, and
max*
2
S
d
are the truncated maximums of
the different sub-distances.
max*
2
max*
2
max
2
max*
2
*
S
S
Q
Q
d
n
d
n
d
n
d
n







max*
2
max*
2
max
2
max*
2
*
S
S
Q
Q
Q
Q
d
n
d
n
d
n
d
n





max*
2
max*
2
max
2
max*
2
*
S
S
Q
Q
S
S
d
n
d
n
d
n
d
n





Notice that, (α,β,γ) depends on the importance of
each type of feature. Firstly, all the components have
the same influence in the calculation of d
2
(i,i’),
because they are proportional to the maximum
distance for each type of feature; secondly, as
truncated maximums are considered they are robust
to outliers; and finally (α,β,γ) are proportional to the
number of features they represent. So, the complete
expression for the compatibility measure is:
)',()',(
)(
)',(
2
2
2
2
2
2
'
),,(
2
iids
n
iid
n
s
xx
iid
Sk
k
S
Qk
k
Q
k
k
kiik











(3)
where
2
k
s
is the variance of the feature X
k
,
)',(
2
iid
k
is
the contribution of a single categorical feature X
k
the
χ
2
measure, and
)',(
2
iik
S
d
is the contribution of the
semantic feature X
k
.
3 THE DATASET
In 2004, the Observatori de la Fundació d’Estudis
Turístics Costa Daurada conducted a study of the
visitors of the Ebre Delta Natural Park. The data
was obtained with a questionnaire made to 975
visitors. The questionnaire was designed in order to
determine the main characteristics of the tourism
demand and the recreational uses of this natural area.
From these questions, two groups of interest
were defined (Anton-Clavé, Nel-lo, and Orellana,
2007): 4 variables that define the tourist profile
(origin, age group, accompanying persons and social
class) and 6 that model the trip profile (previous
planning, first and second reasons for trip,
accommodation, length of stay and loyalty).
3.1 Previous Study and Results
In (Anton-Clavé et al., 2007), an statistical
dimensionality reduction were used to find visitor’s
profiles. In particular, a multivariate homogeneity
analysis was carried out. Two dimensions were
selected for the analysis, keeping a 30% and 26% of
variance respectively. In the interpretation phase, it
was seen that Dimension 1 can discriminate among
the variables relating to type of accommodation,
length of stay and reason for the trip. It shows the
degree of involvement of the tourist with the nature.
The second dimension is determined by the type of
group and age and shows the degree of involvement
with the services available in the park. However, the
total variance represented by the two first
dimensions is 56%, which means that 44% of the
information contained in the data set is missed,
which can seriously affect the interpretation.
From that, five clusters of visitors were identified
in (Anton-Clavé et al., 2007), from which the two
first groups include a total of 83.9 % of the
individuals, concluding that the rest of groups were
really small and targeted to a very reduced group of
visitors. For this reason, only the two main groups,
(EcoTourism and BeachTourism) were characterised
and Chi-square independence test was performed.
In the group EcoTourism (44,6%), the main
interests are nature, observation of wildlife, culture
and sports. People stay mainly in rural
SEMANTIC CLUSTERING BASED ON ONTOLOGIES - An Application to the Study of Visitors in a Natural Reserve
285

establishments and campgrounds. In the group there
are youths (25-24) coming from Catalonia and the
Basque Country and it is the first time that visit the
Delta. The group BeachTourism (39,3%) was
characterised by their interest in beach, relaxation,
and walking. It is family tourism, staying in rental
apartments or second homes. They come from Spain
or overseas. The group contains middle-class people,
35-64 years old, who do long and frequent visits.
4 STUDY OF THE VISITORS
WITH CLUSTERING
In this section, a hierarchical clustering based on the
Ward’s criterion (Ward, 1963) is done on the same
dataset of visitors to the Ebre Delta. We have taken
the same subset of variables (4 that define the tourist
profile and the 6 that model the trip profile).
Table 1: Typology of visitors to the Ebre Delta Natural
Park with categorical features.
Class # Description
864 1 Single outlier visitor
C963 20 Long stage, between 35-early 40s years,
52% stays at second home, 75% are
Catalan people, it is not clear the first
reason to come (some of them come for
walking).
C966 72 Long stage, between 35-early 40s, 68% is
at home, half Spanish, half Catalan, have a
second residence near the park
C965 37 Long stage, higher fidelity, around 46
years, 65% home, 78% Catalan, their main
interest is gastronomy
C921 4 Long stage, more fidelity, between 35-
early 40s, 50% goes to the hotel, an
important part makes reservation, part of
the foreigners concentrated in this group
25% of the class are foreigners, they come
for recommendation of other people, 50%
Catalan, main interests: relaxation or
landscape
C936 16 Shorter stage, between 35-early 40s,
almost 60% home, 80% Catalan, main
interests: nature or business
C918 8 Youngs, under 30s, 50% stay in camping,
75% makes reservation, 50% Spanish,
education tends to be first reason
C964 817 Shorter stage, occasional visit, between 35-
early 40ss, mainly hotel, 63% Catalan,
main reasons: nature, landscape and
sightseeing
In the same way of the previous study, as most of
the variables were categorical, only equal or
different values are distinguished, leading to a poor
estimation of the similarity between responses.
Table 2: Typology of visitors using semantics.
Class # Description
C947 110 The 81% comes for nature, but also for
relax (35%), they use mainly hotels and
rural establishments (79%), they have a
reservation (95%)
C966 194 They come for relax (36%), visit the
family (14.4%), but the second reason is
mainly nature (35%), they have no hotel,
they stay at home or at a family house
(68,5%), and they have no reservation
(99%), this is a group of young people
leaving in the area, which repeat the visits
more than others.
C968 203 Short stage, around 2 days, they clearly
come for nature reasons (91.6%) and
second for relax and wildlife (43.6%), they
are in hotels or apartments (44.6%)
although they have not reservation, mainly
Spanish
C955 88 The first reason for coming is
heterogeneous (nature, relaxation, beach,
landscapes), the second is nature, they stay
in a camping (90%), the half have a
reservation, mainly Catalan and Spanish
but also concentrates a big proportion of
foreigners
C944 124 Relax and wildlife (46%) are the first
reasons for coming and second is nature
(40%), they stay at hotel or cottages
(72%), and have reservation (88%). This is
a group of slightly older people
programming the stay in hotel or
apartment, looking for relax or beach
C964 88 Wildlife and the landscape are the first
reasons for coming (67%), but also for
culture (19.5%) and the second reason is
nature, they are mainly in hotel (54%).
They are mainly Catalan or Spanish.
C957 84 Stay longer, slightly older that the rest,
nature (38%) and beach (16%) are the
main interest and second main interest is
wildlife, most of them are foreigners with
a second home, or that stay in an
apartment.
C961 84 They all come for beach, their secondary
interests are equally relaxation and nature,
they live near the park and their visit is
improvised, the stage is longer.
The clustering generates a big and heterogeneous
class (with 83.8% of the tourists) which seems to
share all type of visitors and other 7 small classes.
Therefore, although the interpretation of the small
classes is possible (see Table 1), from the point of
view of the manager, this partition is useless because
ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence
286

the majority of visitors belong to the same profile as
no clear intra-class difference can be identified.
In the knowledge-based approach proposed in
this paper, the clustering method is able to manage
the meaning of values, relating them to concepts in a
given ontology. Therefore, a semantic clustering has
been done by considering those categorical variables
of the previous experiment as semantic variables and
using the metrics proposed in Section 2.2. WordNet
(Fellbaum, 1998) ontology is used to estimate the
semantic similarity. From the results of this
experiment, a cut in 8 classes is recommended for its
interpretability (Figure 1).
This partition has clusters of more homogeneous
dimension. This is an important fact, since now we
can identify typologies of visitors that represent a
significant proportion of the total number of visitors.
From the dendogram in Figure 1, it can also be
seen that we have obtained clusters with high
cohesion, which means that the distances between
the members of cluster are quite small in comparison
with their distances with objects outside the cluster.
Moreover, if the level of partition is increased, then
the cohesion of the clusters decreases quickly, which
also indicates that the clusters are well defined.
This clustering is coherent with the grouping
made by (Anton-Clavé et al., 2007) using
multivariate analysis, because the variables about the
reasons for visiting the park have a great influence in
the formation of the groups. Interests on nature,
beach and relax are present in different classes.
However, thanks to the semantic interpretation of
the concrete textual values provided by the
respondents, we have been able to identify that
visitors interested in nature are similar to those
interested in wildlife. The system has been also able
to identify the similarity between hotels and cottages
and between second homes and familiar houses.
This proofs that the estimation of the relative
similarities among objects in terms of the meaning
of the values improves the final grouping.
In this way, the two types of visitors identified in
statistical analysis as Ecotourism and Beach
Tourism have now been refined as follows:
 Ecotourism: visitors that stay in hotels and
apartments for relax (C947), visitors with familiars
or a second residence (C966), Spanish visitors
interested in wildlife (C968) and tourists interested
in culture (C955).
 Beach tourism: older people staying in hotels or
apartments looking for relax and people that live
near the park and go to the beach.
Notice that this is a more rich classification that
establishes clear profiles of visitors.
5 CONCLUSIONS
The exploitation of data from a semantic point of
view establishes a new setting for data mining
methods. In this paper, it has been proposed a
method to include semantic variables into an
unsupervised clustering algorithm. A combination
function that combines numerical, categorical and
semantic features has been formally defined. In
particular, the contribution of semantic features is
obtained by estimating the semantic similarity of
textual values from a conceptual point of view
exploiting ontologies. Then, a knowledge-based
clustering is proposed.
The paper presents an application of this
methodology to a dataset obtained from a survey
done to the visitors of a Natural Protected Park. The
results show that a semantic clustering approach is
able to provide a partition of objects that considers
the meaning of the textual responses and, thus, the
result is more interpretable and permits to discover
semantic relations between the objects. The method
has produced a more equilibrated grouping and
provides useful knowledge about the characteristics
of the visitors.
After obtaining these promising results, we will
study the effect of using domain ontologies (e.g.
medicine) instead of WordNet in the semantic
similarity assessment. Moreover, the consideration
of a set of ontologies used in an integrated way is
also under study.
ACKNOWLEDGEMENTS
This work has been partially supported by the Universitat
Rovira i Virgili (2009AIRE-04) and the DAMASK
Spanish project (Data mining algorithms with semantic
knowledge, TIN2009-11005) and the Spanish Government
(PlanE, Spanish Economy and Employment Stimulation
Plan). M. Batet is supported by an URV research grant.
SEMANTIC CLUSTERING BASED ON ONTOLOGIES - An Application to the Study of Visitors in a Natural Reserve
287


Figure 1: Dendogram with semantic features (8 classes).
REFERENCES
Ahmad, A., and Dey, L. (2007). A k-mean clustering
algorithm for mixed numeric and categorical data.
Data Knowledge Engineering, 63(2), 503-527.
Anton-Clavé, S., Nel-lo, M.-G., and Orellana, A. (2007).
Coastal tourism in Natural Parks. An analysis of
demand profiles and recreational uses in coastal
protected natural areas. Revista Turismo &
Desenvolvimento, 7-8, 69-81.
Batet, M., Sanchez, D., Valls, A., and Gibert, K. (2010).
Exploiting Taxonomical Knowledge to Compute
Semantic Similarity: An Evaluation in the Biomedical
Domain. In Trends in Applied Intelligent Systems.
23rd International Conference on Industrial
Engineering and Other Applications of Applied
Intelligent Systems, IEA/AIE 2010, LNAI 6096 (pp.
274-283): Springer.
ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence
288

Fellbaum, C. (1998). WordNet: An Electronic Lexical
Database. Cambridge, Massachusetts: MIT Press.
More information: http://wordnet.princeton.edu.
Gibert, K., and Cortés, U. (1997). Weighing quantitative
and qualitative variables in clustering methods.
Mathware and Soft Computing, 4(3), 251-266.
Leacock, C., and Chodorow, M. (1998). Combining local
context and WordNet similarity for word sense
identification. In WordNet: An electronic lexical
database (pp. 265-283): MIT Press.
Lin, D. (1998, July 24-27). An Information-Theoretic
Definition of Similarity. Paper presented at the 15th
International Conference on Machine Learning
(ICML98), Madison, Wisconsin, USA.
Rada, R., Mili, H., Bichnell, E., and Blettner, M. (1989).
Development and application of a metric on semantic
nets. IEEE Transactions on Systems, Man, and
Cybernetics, 9(1), 17-30.
Resnik, P. (1995, August 20 - 25). Using Information
Content to Evalutate Semantic Similarity in a
Taxonomy. Paper presented at the 14th International
Joint Conference on Artificial Intelligence, IJCAI
1995, Montreal, Quebec, Canada.
Studer, R., Benjamins, V. R., and Fensel, D. (1998).
Knowledge Engineering: Principles and Methods.
Data and Knowledge Engineering, 25(1-2)(1-2), 161-
197.
Ward, J. H. (1963). Hierarchical Grouping to Optimize an
Objective Function. Journal of the American
Statistical Association, 58, 236-244.
SEMANTIC CLUSTERING BASED ON ONTOLOGIES - An Application to the Study of Visitors in a Natural Reserve
289