Mina Ziani, Danielle Boulanger and Guilaine Talens
University of Lyon – Jean Moulin, Modeme Team, 6 Cours Albert Thomas, Lyon, France
Keywords: Ontology alignment, Ontology integration, Hybrid ontology, Mappings, Semantic interoperability.
Abstract: In the context of the cooperation between heterogeneous and distributed information sources, ontologies are
a main issue. To represent shared knowledge, an hybrid domain ontology is designed and to respect each
point of view of different experts, local ontologies are created. Since experts are willing to cooperate,
similarities must be identified to build mappings between the concepts of the different ontologies. We
propose a computer-aided system to allow the experts to choice similarity measures on demand. We apply
this work to the geotechnic domain which involves various businesses.
The knowledge management in the context of
heterogeneous and distributed information sources is
a great challenge. The difficulty is to represent all
the domain knowledge specificities and to allow the
cooperation between experts with different points of
view. Ontologies are a promised approach for the
knowledge representation in a formal way.
We propose in this paper an approach to
establish interoperability between knowledge
contained in different local ontologies. An hybrid
ontology (Visser, 2002) is designed. It consists to
describe each information source in a local ontology
and to represent the vocabulary shared by all the
Each expert community built her own business
ontology. A global ontology is automatically
designed and contains only the common concepts
and properties of all the local ontologies.
To reach cooperation between the several
ontologies and to allow semantic interoperability,
different techniques are used. In particular, ontology
alignment, or matching, is the process of
determining relationships or correspondences
(subsumption, inclusion …) between entities of
different ontologies. The correspondences are also
called alignments. Very often, these relations are
“equivalence relations” discovered through
similarity measures between ontologies’ entities. In
the existing alignment system, these measures are
used, alone or aggregated according to a particular
strategy. This one depends to the domain, the type,
and the use of the ontology. So, we have developed
a generic computer-aided system which guides the
experts to choice the similarity methods and
measures to be used in the alignment process
according to the ontology characteristics.
We applied this work to the geotechnics. It is a
complex domain involving different businesses:
Project management, geologists and chemists…To
design ontology for this domain we use information
contained in the geotechnic referentials and
We present some tools and frameworks for
merging or aligning ontologies. After, we describe
the hybrid ontology approach to represent
geotechnical knowledge. In order to allow
cooperation between experts, we propose a
framework enable to create mappings. Finally, we
conclude with some perspectives.
In OntoDB project, each information source builds
its local ontology from the concepts and relations
contained in a global domain ontology. So, semantic
integration is automatic. This research work utilizes
a strong hypothesis: A database administrator
defines a relevant ontology and he adds the
subsumption relations existing between his local
ontology and the global ontology (Nguyen Xuan,
OWSICS architecture involves two ontology levels:
Ziani M., Boulanger D. and Talens G..
DOI: 10.5220/0003651802140219
In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2011), pages 214-219
ISBN: 978-989-8425-80-5
2011 SCITEPRESS (Science and Technology Publications, Lda.)
The local information sources and the cooperation
(global). Information sources are semantically
explained with corresponding local ontologies. At
the global level reference ontology describes the
domain semantics. A semi-automatic method has
been developed to allow the creation of mappings,
but only between a local ontology and the reference
ontology (Abrouk, 2008).
Except OntoDB, cooperation systems need
ontology alignment. Then, we focus on several tools
to align ontologies. Most of them use terminological,
conceptual or extensional similarity measures and
combine them according to an aggregation strategy.
They differ in their functioning and the interactions
they offer to the users.
PROMPT is a computer-aided system for
comparing, merging, aligning and managing
ontologies. Its alignment and merge module, called
Anchor PROMPT, allows the expert, to find
mapping in the following way: (i) the system
calculates terminological measures to determine an
initial set of similar concepts, (ii) from this list, an
algorithm analyzes paths in the sub-graphs bounded
by these concepts and indicates which classes
frequently appear in the same positions on similar
paths (Noy, 2001).
OLA (OWL Lite Alignment) is a system
implementing an algorithm of ontology alignment
written in OWL. It measures the similarity between
two entities from the similarity between their
characteristics (classes, properties, relations with the
other entities…). The final similarity is the weighted
sum of these similarities. The weights are associated
relatively to the type of entities to be compared. The
algorithm uses a fixed point method with iterations
to improve the similarity of two entities. When there
are no possible improvements, alignments between
two ontologies are proposed (Euzenat, 2004).
AROMA (Rule Ontology Matching Approach
association) is an approach of alignment of
ontologies represented in OWL. It allows
discovering semantic links (subsumption or
equivalence between two entities: Classes or
properties). There are three steps in the process of
alignment: The first one consists in the acquisition
of “relevant terms” for each concept and its
ancestors. These terms, contained in the descriptions
and instances of ontology entities, are extracted with
tools of Natural Language Processing. The second
step allows creating relations of subsumption
between the ontology entities from rules of
association. In the final step, the system analyzes the
relations previously obtained in order to: (i) deduce
the relations of equivalence; (ii) find inconsistences
and eliminate them (iii) delete the redundant
relations; (iv) select the best alignment for every
entity (David, 2009).
More recently, frameworks have appeared in the
ontology alignment systems. They allow multiple
combinations of strategies to calculate the similarity.
For example:
COMA++ (COmbining MAtching) is a generic
system of matching schemas (Do, 2002; Massmannn
2006). This framework allows the importation,
storage, and edition of schemas and produces
alignment algorithms in order to transform or merge
those schemas. It provides an extensible library of
alignments, a module for the combination of the
results and a platform to estimate the various
measures. The user can interact during the matching
process by selecting the measures aggregation
strategy (the average, the weighted sum,…).
MAFRA (MApping FRAmework for distributed
ontologies) is an interactive framework, dynamic
and progressive, for the alignment of ontologies
distributed through the semantic web (Mädche,
2002). The steps of the MAFRA alignment process
are: (i) Importation and standardization of ontologies
to align, (ii) similarities to compute between
elements of different ontologies from a combination
of similarity measures, (iii) formalization of
mappings by establishing "semantic bridges"
between the entities of different ontologies, (iv)
execution of mappings to transform the instances of
source ontology to the instances of a target ontology
based on semantic bridges, and finally the results’
FOAM (Framework for Ontology Alignment and
Mapping) is a framework used in several systems for
data integration, ontology merging, and ontology
evolutions... The tool implements several measures
and strategies for similarities research and allows to
create mappings between ontologies described in
OWL. The general process of alignment is as
follows: One selects the pairs of entities to be
compared and the characteristics on which to
perform the comparison. The system calculates a
similarity for each pair and for each characteristic.
These results are combined to obtain the final
similarity between each pair of entities. From these
results, FOAM forwards a set of suggestions for
alignment, to be validated by the users (Ehring,
RiMOM (Risk Minimisation based Ontology
Mapping) is an interactive framework which
implements several strategies to align ontologies
(Tang, 2006); (Li, 2009). The RiMOM’s alignment
process is as follows: The first step consists in
selecting the used measures depending on the
assumed similarity between the ontologies
(terminological or structural). In the second step,
several measures are applied. Then, the results are
combined using linear interpolation function. The
third step is the propagation of the similarities (from
concept to concept, from property to property, from
concept to property). The final step consists in
generating mappings from the results previously
obtained. The process is iterative, with a validation
of results at each iteration.
Tools and frameworks we presented differ in the
measures used and the aggregation strategy. These
systems do not guide user, according to the context,
to use a particular methods to discover similarity. In
addition, these systems do not permit to reuse
previously calculated result.
In order to improve the interactions with the
users, we design a computer-aided system allowed
selected similarity methods according to the
characteristics of the concepts to align. The expert is
guided in the selection of the methods to be
aggregated. Our contribution is to propose a
computer-aided system which calculates similarities
and discovers semantic relations. Similarities and
mappings are stored in databases and can be reused
in order to avoid new computation.
Geotechnics is the science studying the grounds
according to diverse aspects: Mechanics of grounds,
geology, techniques of building…The complexity of
geotechnical domain and the knowledge
heterogeneity imply the sharing and the management
of knowledge to be difficult (Faure, 2007).
To represent knowledge of geotechnical domain,
ontologies are a main issue. The existing concepts in
the domain are too numerous to be represented in a
single ontology (approximately 5000 concepts), so
we propose to design an hybrid ontology.
Each group of experts builds an ontology
representing concepts, properties and instances of
his business: A local ontology. A global ontology is
automatically created by the system to represent a
shared vocabulary.
Initially, we classify the ontologies by level. The
first level corresponds to the ontology sharing the
most common concepts with the others. It represents
the target ontology. The source ontologies are the
other local ontologies.
The concepts and properties of the source
ontology involved in the target ontology are
integrated to the global ontology in the ascending
order of ontology levels.
The result is an ontology represented by a
conceptual graph. This one is verified by the means
of an algorithm of integration (Ziani, 2008) which
allows the deletion of the relations providing cycles
in the global ontology. These last ones are stored in
the local ontologies. Finally, we obtain a consistent
and consensual ontology including the common
concepts and properties of local ontologies and not
the conflicting relations which connect them.
Let a target ontology “Tunnel” containing the
concepts and instances used in tunnel engineering
and a source ontology “Project management”
containing the concepts and instances used by the
experts in project management (cf. figure 1).
Figure 1: Description of the ontologies “Tunnel” and
“Project management”.
The integration program allows deleting the
relation directed from “structure” to “tunnel” in the
global ontology, because there is a cycle between
these two concepts and the relation between them is
not in the target ontology “Tunnel”. This relation is
preserved in the source ontology “Project
management” (cf. figure 2).
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
Figure 2: Global ontology.
This representation is possible relatively to the
plurality of the geotechnic sub-domains and the
existence of a vocabulary shared by all the experts.
The hybrid ontology contains ten business
ontologies (currently, we have pointed ten differents
sub-domains) and a global one, written in the OWL
Each local ontology is represented as a tree,
simple to implement and destinated to geotechnical
experts. Its size can vary according to the business
(between 100 and 1000 concepts).
To permit cooperation business ontologies, we
propose a computer aided system (MOON). It
guides the expert in the process of mapping creation
between concepts of local ontologies (Ziani, 2011).
The architecture of this system is presented in figure
Most of the alignment systems use various
measures of similarity to deduce the similarity
between two entities. The difficulty is to choose the
right measure or the combination of measures to find
the similarity. Our system helps the geotechnical
expert in the process of similarity research between
concepts and generates mappings between them.
When an expert wants to cooperate with another,
he sends to the system a request including two
ontologies: Departure (corresponding to the business
of the expert) and research and, the concept to align
(departure concept) (1). The system loads the two
local and the global ontologies (2). The objective is
to discover the concepts of the research ontology to
align with the concept of the departure ontology.
Then, the system verifies in the similarity database if
there are synonyms for the departure concept (3).
Several methods of similarity measures are
implemented in the framework and can be proposed
to the expert (4). The interest of the framework is to
reuse different implemented measures
(terminological, conceptual...) and to allow the
combination of several similarity measures. The goal
of the system is to give to the geotechnical experts
several measures and to help them choosing the best
The expert selects the methods and measures (5).
The result is a set of similarities between the
departure concept and the concepts found in the
research ontology. These can be of different types
(equivalence, subsumption) and are stored in the
similarity database (6). Then, they are proposed to
the domain expert (7). The expert can validate or not
these similarities. The semantic links proposed and
the expert names are stored in the mapping database
(8). If the expert knows the concepts to
align, he
directly stores the relation in the mapping database.
This involved the update of the generated mappings
Figure 3: Architecture of the MOON system.
When an alignment, between two concepts, is
proposed, the system researches in the mapping
database a relation between the same concepts. If
there is no relation between the concepts, this one is
automatically generated in the ontologies.
Otherwise, there are two possibilities: Either the
same alignment exists, in this case there is no
modification to be brought to the ontologies, or there
is a contradictory alignment: In this case, we cannot
create this latter. The alignment previously created is
deleted. It can be recreated only by a third expert
who confirms one of the existing solutions or by an
expert who modifies the alignment which he has
previously proposed in the mapping database.
The process to guide the experts to choose
similarity measures is explained in the following
At first, the system calculates the terminological
similarity measures between the departure concept
and all the concepts of the research ontology. When
the terminological similarity measure gives a result,
it means the compared concepts are lexically similar.
After, the system verifies if the concepts are
semantically similar from the global ontology. It
researches and compares the smallest subsuming
concept of the departure concept and the found
concept which exists in the global ontology. If they
are identical, they are considered as potentially
similar. On the contrary, if a link of subsumption in
the global ontology exists, the system deduces a
similarity between them. But if they are no relation
between them in the global ontology, the system
deduces that the compared concepts are not
semantically similar.
Then, if the departure concept contains at least
two attributes, the system proposes to calculate the
similarity measure based on the concept properties.
This measure gives the concepts in ontology
research which have common attributes with the
departure concept.
Then, the system proposes a method based on the
hierarchical structure of ontology: Counting the
edges. It consists in (i) researching the smallest
subsuming concepts of the departure which exists in
the global ontology, (ii) calculating the number of
edges between this common concept and the
departure concept, (iii) selecting all the concepts in
the research ontology from the level N-2 of the
common concept until the level N+2, (iv) suggesting
to the expert the found concepts if their number is
not very important (subordinate or equal to 10).
In addition, to improve these results, the system
proposes extensional methods: Two classes are
similar if they share a subset of instances for
attributes chosen by the expert.
Suppose that the “Project management” ontology
has to cooperate with “Tunnel” ontology. The
discovery of mappings through the system for the
concepts “digging” and “crane” gives the following
For the concept "digging", the terminological
similarity measures between concepts do not give
result. The system calculates the similarity measure
based on the attributes. It finds the same attributes in
the concept “digging” and the concept “earthwork”
containing in the “Tunnel” ontology: “method” and
“ground quantity”. Therefore, it suggests to the
expert to create the relations between them.
For the concept “crane”, the terminological
similarity measures find the concept “truck crane”.
Similarity based on the attributes is not proposed
because the concept “crane” contains less than 2
attributes. Consequently, the system proposes the
method of the edge counting: It researches the
smallest concept subsuming this concept and
existing in the global ontology. It finds “construction
site”. There are 2 edges separating these two
concepts, so it selects all the concepts in the
“Tunnel” ontology from the level 0 (level of the
concept “construction site”) to 4 (4
level from the
concept “construction site”). They find the concepts:
“construction site”, “construction tools”, “truck
crane”, “sealing material” and “tunnel boring
machine”. The number of these concepts is less than
10, the system proposes them as similar concepts
with priority to the concept “truck crane” finding
This work offers a guide to an expert in the
process of creating mappings. All the result found
must be stored in the similarity and mapping
In this article we have presented an approach for the
representation of heterogeneous, distributed
knowledge and the cooperation between experts of
various businesses. We applied this work to the
geotechnical domain.
An hybrid ontology is designed: In the first time,
the business ontologies are conceptualized by a
consensus between several experts in the same sub-
domain. At the end, the global ontology is
automatically built from these local ontologies by an
integration approach. To allow the cooperation
between the experts, we have implemented a
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
prototype with a module to guide experts to align
ontologies. In current systems the measure scheduling
is fixed; on the contrary, in our proposition, the
system selects the measures relatively to the ontology
A similarity database stores all the calculated
similarities and a mapping database containing all the
relations validated by the experts. All the stored
measures can be reused to avoid new computations
and so not to perform the process.
Currently, the system of the geotechnical
knowledge management is partially implemented.
The local ontologies and the created mappings
between concepts evolve. They imply modifications
in the global ontology and the generated mappings.
So, the first perspective of this work is to analyze the
consequences of the hybrid ontology evolution and
to propose some solutions to maintain the
consistence of all the ontologies (local and global).
There are diverse systems which manage the
ontology evolution (Stojanovic, 2004; Jaziri, 2010;
Djedidi, 2010). Our future contribution will manage
an hybrid ontology evolution.
The second perspective is to estimate all the
mappings stored in the similarity database. The
interest is to deduce other semantic relations.
Finally, the third perspective is to study the
scalability of the hybrid ontology and the alignments
between concepts.
Abrouk L., Cullot N., Ghawi R., Gomez Carpio G-V.,
Poulain T., 2008. Cooperation of information sources
in OWSCIS System, In CGCT’2008.
David J., 2009. AROMA results for OAEI 2009, In
OM’2009, 4th ISWC workshop on ontology matching,
p.147-152, 2009.
Djedidi R., Aufaure M-A. ONTO-EVO
L an Ontology
Evolution Approach Guided by Pattern Modeling and
Quality Evaluation, In FoIKS’2010, p. 286-305.
Do H., Rahm E, 2002. COMA – a system for flexible
combination of schema matching approaches, In
VLDB’02, 28
International Conference on Very
Large Data Bases, p. 610-621.
Ehring M., 2007. Ontology Alignment: Bridging the
Semantic Gap: Semantic Web and Beyond, Springer.
Euzenat J., Valtchev P., 2004. Similarity-based ontology
alignment in OWL-lite, In ECAI’04, 15
Conference on Artificial Intelligence, p. 333-337.
Faure N., 2007. Un système d’aide à la modélisation des
connaissances en géotechnique, Thèse de doctorat en
informatique, Université Jean Moulin – Lyon3.
Jaziri W., Sassi N., Gargouri F., 2010. Approach and tool
to evolve ontology and maintain its coherence,
Metadata, Semantic and Ontologies, vol. 5, n°2,
p. 151-166.
Li J., Tang J., Luo Q., 2009. RiMOM: A Dynamic
Multistrategy Ontology Alignment Framework, IEEE
Transactions on Knowledge and Data Engineering,
vol. 21, n°8, p. 1218-1232.
Mädche A., Motik B., Silva N., Volz R., 2002. MAFRA-a
MApping FRAmework for distributed ontologies, In
EKAW’02, International Conference on Knowledge
Engineering and Knowledge Management, Springer,
p. 235-250.
Massmann S., Engmann D., Rahm E., 2006. COMA++:
Results for the Ontology Alignment Contest OAEI, In
OM’2006, 4th ISWC workshop on ontology matching.
Nguyen Xuan D., Bellatreche L., Pierra G., 2006. A
Versioning Management Model for Ontology-Based
Data Warehouses», DaWaK 2006, vol. 4081, 2006, p.
Noy N., Musen M., 2001. Anchor PROMPT: Using non-
local context for semantic matching, In IJCAI’01,
Workshop on Ontology and Information Sharing,
p. 63-70.
Stojanovic L., 2004. Methods and Tools for Ontology
Evolution, PhD thesis, University of Karlsruhe.
Tang J., Li J., Liang B., Huang X., Li Y., Wang K., 2006.
Using bayesian decision for ontology mapping,
Journal of Web Semantics, vol. 4, n° 4, p.243-262.
Visser U., Stuckenschmidt H., Schlieder C., Wache H.,
Timm I., 2002. Terminology Integration for the
management of distributed information resources,
Kuenstliche Intelligenz Journal, vol. 16, p. 31-34.
Ziani M., Boulanger D., Talens G., 2008. Designing an
Hybrid ontology From Domain Ontologies, In
JFO’08, 2
Journées Francophones sur les
Ontologies, Lyon, p. 41-47.
Ziani M., Boulanger D., Talens G., 2011. Système d’aide à
l’alignement d’ontologies métier–Application au
domaine de la géotechnique, revue Ingénierie des
systèmes d’informations, vol. 16, n°1, p. 89-112.