=Paper=
{{Paper
|id=Vol-292/paper-4
|storemode=property
|title=Understanding and Supporting Ontology Evolution by Observing the WWW Conference
|pdfUrl=https://ceur-ws.org/Vol-292/paper4.pdf
|volume=Vol-292
|dblpUrl=https://dblp.org/rec/conf/semweb/GuelfiPR07
}}
==Understanding and Supporting Ontology Evolution by Observing the WWW Conference==
https://ceur-ws.org/Vol-292/paper4.pdf
Understanding and Supporting Ontology Evolution by
Observing the WWW Conference
Nicolas Guelfi1, Cédric Pruski1,2, Chantal Reynaud2
1
Laboratory for Advanced Software Systems – University of Luxembourg campus
Kirchberg 6, rue Coudenhove-Kalergi, L-1359 Luxembourg
2
LRI – University of Paris-Sud, CNRS & INRIA-Futurs, 4 rue Jacques Monod, Parc Club
Orsay Université, 91893 Orsay Cedex (France)
{nicolas.guelfi, cedric.pruski}@uni.lu, chantal.reynaud@lri.fr
Abstract. Ontologies which represent domain knowledge in information
systems are efficient to enhance information retrieval. However, domain
knowledge is evolving over time and thus it should be also expressible at
ontology level. Unfortunately, we consider that ontology evolution is barely
study and its basic principles have not been yet precisely defined according to
our notion of evolution. In this paper, we have followed a bottom-up approach
consisting in a rigorous analysis of the evolution of a particular domain over a
significant period of time (namely the WWW series of conference over a
decade) to highlight concrete domain knowledge evolutions. We then have
generalized and we present a precise set of evolution features that should be
offered by ontology metamodels. We also evaluate the modelling capabilities of
OWL to represent these features and finally, we show the contribution of
ontology evolution support to improve Web information retrieval.
Keywords: Ontology Evolution, Domain Analysis, OWL, Web Information
Retrieval
1 Introduction
Although being firstly introduced in philosophy, ontologies have recently appeared in
the field of computer science as the cornerstone of the Semantic Web paradigm [3].
The later aims at giving a sense to the Web what will allow computers to
“understand” Web data. If this goal is achieved, computers will be able to unload
users of many tedious tasks like searching relevant documents or services. The
Semantic Web implements ontology that models a part of the real human world,
mainly to annotate Web data or to facilitate information retrieval. Nevertheless, since
ontologies represent the knowledge of a particular domain, they have to smoothly
follow the evolution of that domain otherwise their use will lead to unwanted effects.
Therefore the ontology evolution problem [11], [15] has recently been deep studied
since it becomes rapidly of utmost importance.
ESOE, Busan - Korea, November 2007 19
In our previous work we have defined the O4 approach [6], [8] that aims at
improving the results of a Web search in terms of relevance. This is done mainly
through the use of ontology-based query expansion rules. In order to optimize the
search, we need to select the adequate terms from the ontology to enrich the query.
Actually, if the ontology does not reflect the knowledge of the domain associated to
the submitted query, the search results will not be those awaited by users. We are thus
facing the problem of ontology evolution.
In this paper we propose a set of modelling features for ontology evolution. These
features have been defined after the rigorous study of the evolution of a particular
domain (in our case, the domain defined by the WWW series of conference topics)
over a ten years period of time. In consequence, we will first have to present in detail
the construction of a corpus of documents that is representative of the domain we
have studied. This requires the definition of relevant criterion and tools that will
facilitate the analysis of the domain. The results of this analysis will lead directly to
the definition of the various kind of evolution that can appear [7] which in turn will
allow the proposition of modelling features that aims at designing evolving
ontologies. The proposed primitives will first allow us to understand the evolution of
ontologies and will aid to predict future versions of ontologies. They can be used to
describe a structural evolution on one hand and a progressive evolution on the other
hand. Since this work has been carried out in a context covering Web information
retrieval, we will highlight the contribution of such ontologies through an example
implementing ontology-based query expansion techniques [8] to improve the
relevance of documents when searching the Web.
The remainder of the paper is structured as follows. Section 2 presents the
characteristics of the domain we have studied in order to define the new modelling
features devoted to ontology evolution. In Section 3 we detail the proposed modelling
primitives as well as their properties. Section 4 illustrates an application of our work
through a basic example dealing with Web information retrieval. In Section 5 we
discuss related work in the ontology evolution field. Finally the paper wraps up with
concluding remarks and our future work.
2 Domain of Study Definition and Ontologies Construction
The first step towards the proposition of modelling features devoted to ontology
evolution concerns the construction of a significant corpus of documents that will
allow us to highlight the various kinds of domain evolution. In this section we present
the characteristics of such a corpus and the ontologies we have built from that pool of
documents.
2.1 Domain Selection
Since we want to derive modelling features for ontology evolution from the analysis
of the evolution of a particular domain, the selection of such domain is of utmost
importance. Many domains, like bioinformatics through the Gene Ontology [16], are
already modelled using ontologies. Unfortunately, these ontologies are either young
20 International Workshop on Emergent Semantics and Ontology Evolution
or built using only domain-specific relations. Therefore we considered that the study
of their evolution will not be relevant enough and we decided to construct ontologies
from a set of descriptions of an evolutionary domain. To this end, we have chosen the
domain covered by the World Wide Web series of conference which is reflected in
the calls for papers and in the accepted papers. Thus, papers accepted for publication
at these events together with the calls for papers, which are online and can be
retrieved via a web search engine, form our “Micro Web” (i.e. the corpus of
documents we will analyze). In order to see the evolution of the domain of the Web
over a significant period of time, we decided to harvest the accepted papers of the last
10 WWW events. The so-called Micro Web consists of good case study since the
chosen conference is world famous and known to be one of the most representative
events in the domain of the Web. Therefore, its successive calls for papers reflect the
various fields in vogue in the corresponding domain. Moreover, the quality, quantity
and homogeneity of the submitted papers as well as the high level of selectivity (less
than 20%) set by reviewers reinforce this idea. As a result, we have a corpus made of
622 documents all stored in a relational database which will facilitate their future
analysis.
2.2 Methodology for Ontology Construction
We have designed, from the calls for papers and the accepted papers of the different
conferences, the ontology of the domain of the corresponding event for each year
using the Protégé 1 ontology editor. It means that we have built 10 ontologies, one
concerning each event of the WWW series. These various ontologies represent in fact
the same domain that is evolving over time. The ontologies are constructed following
a rigorous process inspired from the ARCHONTE methodology [2]. The three steps
of this methodology consist in a semantic normalization of the terms introduced in the
ontology, followed by a formalization of the meaning of the knowledge primitives
obtained and an operationalization using knowledge representation languages. The so
built ontologies will allow us to identify the different evolutions of the domain to try,
in the next phase, to explain the changes. The construction of the different ontologies
has been done manually following the process described hereafter. We first model the
knowledge of the domain and then we formalized it using the Web Ontology
Language (OWL). As modeling is the main purpose here, we use the most expressive
flavor of OWL (i.e. OWL Full).
First of all, we stated that every topic of a call for papers denotes a concept in the
corresponding ontology. This means that our ontologies are small and made of about
30 classes. For instance, the topic multimedia in 1998 provides the concept
multimedia in 1998’s ontology. Furthermore, topic like social and cultural gives rise
to two concepts (i.e. a concept social and another one cultural) in the ontology.
Indeed, we decided to split expressions according to the conjunction “and” which
regularly appears. Nevertheless, the conjunction of words using this proposition
indicates that words involved in that particular topic are linked. This is the first step
towards the construction of the set of relations that bind concepts of the ontology.
1
http://protege.stanford.edu/
ESOE, Busan - Korea, November 2007 21
Then, to determine these relations, we base on the content of the accepted papers
with a particular attention devoted to the abstract of the papers. We first localize an
occurrence of the concepts we tried to bind in a document of our corpus, and then we
tried to identify manually from the text the relation of the domain. In order to validate
our choice, we reiterated the operation on several other papers. This basic but rigorous
process provides us the ontologies (one is depicted in figure 1) of the domain that will
be the material of our study devoted to domain evolution.
Fig. 1. Part of the Ontology representing WWW 2000’s call for papers
As partially illustrated on the ontology in figure 1, we used very elementary
relations like subsumption, equivalence or meronymy (i.e. part-of) to design our
ontologies. Nevertheless, we also introduce some particular relations resulting from
the analysis of the content of the papers like use or allow. All the constructed
ontologies can be downloaded from our Web site 2 .
3 Domain Analysis: Towards Ontology Evolution
The analysis of the evolution of the domain represented by WWW conferences’
topics made it possible to define various kinds of evolution that affect the domain. In
this section, we will detail these evolutions and as a result, we will give the
corresponding modelling primitives for the design of evolving ontologies.
3.1 Domain Evolution
In this first subsection, we discuss the various kind of evolution that stand out during
the analysis phase of our micro web. The analysis we made is at two levels. The first
one, called general observation, defines the macroscopic evolution of the domain over
a long period of time (10 years in our case). In contrary, the second level, called local
observation, highlights the microscopic variation of the domain. This second kind of
observations is made on a very short period (i.e. 1 or 2 years). These observations
2
http://se2c.uni.lu/tiki/tiki-index.php?page=TargetTool
22 International Workshop on Emergent Semantics and Ontology Evolution
made it possible to emphasis different kinds of evolution among which one finds
concept persistence, emergence of new concepts, concepts removal, generalization
and specialization of concepts. However, our observations have also permitted to
define other important features like the importance of concepts in the domain but also
the resistance to modification and the variation of a distance between concepts over
time as well as the speed of evolution. All these characteristics will be detailed in the
remainder of this section. However, in order to explain the various highlighted kinds
of evolution, we needed to distinguish between the ontology built from the calls for
papers which represent the conferences chairman point of view and the content of the
papers which represent the authors’ interpretation of the calls. Unfortunately, we did
not have access to the reviews. These would have permitted to understand if the
authors’ interpretations of the calls were consistent with the chairs point of view of
the domain.
Concept Persistence
This first kind of evolution affects some particular concepts of the domain. Actually,
we observe that special concepts like security or search are present in the ontology
over the whole period of observation. It means that since its appearance in the
domain, the concept remains in the domain. We called this constraint on evolution
concept persistence. Our personal knowledge of the domain lets us claim that these
two concepts denote key notions of the domain (recall that we study the domain
covered by World Wide Web series of conference). Thus, we can say that concepts
that are part of the ontology over a predefined long period of time constitute the core
of the ontology because the semantics of the concept is still covered by the semantics
of the domain. This is particularity important for approach exploiting ontologies like
techniques for indexing data or data retrieval. In fact, these concepts are the most
relevant and in consequence should be favoured in their usage. For instance, the
search concept is present in the domain for the whole period of study, whereas other
concepts which seem to be less important like social remains in the domain only for
one year. We will illustrate this particular point in Section 4 hereafter.
Concept Emergence
The second observation in the evolution of a domain concerns the addition of
knowledge at a particular moment. This emergence of concept was particularly true
for the Semantic Web in 2002. Since this paradigm was defined in 2001 by Tim
Berners-Lee [3], and its associated semantics was close enough from the semantics
covered by the domain defined by the topics of the WWW series of conference, it
rapidly appeared as a concept of the domain and as a result, one year later in the
topics of the WWW conference. This is why it takes place in our ontology
representing the domain covered by WWW 2002 topics. Our survey has shown that
79 concepts have emerged in the domain of the WWW series of conference between
1997 and 2007. Moreover, there are about 11 concepts in average that emerge each
year in an ontology that contains about 30 concepts. Recall that we have one ontology
per conference.
ESOE, Busan - Korea, November 2007 23
Concept Removal
A concept can be removed from a domain for several reasons. The first one is related
to its semantics. Actually, by virtue of knowledge evolution, the semantics of that
concept could not be covered by the semantics of the domain (described by the
ontology) anymore and therefore should be removed from that domain. Moreover, a
concept can be either not precise enough (i.e. the concept is too abstract) or too
precise (i.e. specific concept). This would also require some domain refinement which
in turns will lead to the removal of concepts for the benefit of more abstract or
specific concepts. Another reason concerns the properties of the concept. For
instance, if the concept is no more “popular” or “profitable” (if we are in a business
domain) for the domain it can also be removed. We can speak about obsolete concept.
In our case study this kind of evolution arose several times. For instance, concepts
like social and cultural appear in the 1998 WWW conference topics but are removed
in the 1999 conference topics and does not appear anymore. Moreover, our study
revealed that in the WWW 1998 conference, no papers containing these two words
were submitted which proves a kind of irrelevance. This is probably the reason why
both concepts have been removed from the domain from that moment.
Concept Abstraction
Our observations revealed that a concept or several concepts can be substituted along
time axis by a more general concept. We call this phenomenon concept abstraction.
This can be done when the semantics of a concept is completely covered by the
semantics of a concept that is directly link to it. However, we observe that this
phenomenon usually turns up when a concept is becoming less relevant for the
domain. For instance, in our ontologies concepts like browser and tool are generalized
into the more general concept application. This substitution give less importance to
the two concepts that have been generalized which in turns give more freedom for the
future authors in their interpretation of the call for papers. Actually, since this
evolution in the call for papers took place, there have been more submitted papers
dealing with a wider range of applications than papers discussing only the use of Web
browsers. Our study has permitted to emphasis this idea. Concretely, there are 351
occurrences of the word browser, 173 occurrences of the word tool and 351
occurrences of the term application in the documents. Moreover, 30% of these words
are cited in the same papers and in most of the cases, the words browser and tool can
be replaced by the term application without a loss of semantics (i.e. the sentences
where this phenomenon appears have the same meaning after terms substitution).
Therefore the concepts of the ontology representing these notions (i.e. browser and
tool) have been substituted by a more general one (i.e. application) which gave place
to a wider variety of papers on Web applications. We have identified 5 concepts that
have been abstracted. However, the time needed for a concept to become more
abstract varies. Actually, some abstractions are very fast, only one year for the
abstraction of concepts like browser and tool, other highlighted abstractions can be
longer.
24 International Workshop on Emergent Semantics and Ontology Evolution
Concept Specialization
In the contrary, our empirical study has shown that a concept or a group of concepts
can evolve in a more specific concept. Contrary to concept abstraction presented
above, this phenomenon is possible only if the more specific concept on one hand
shares a part of the semantics of its super concept and on the other hand, offers some
specifics axioms that make it possible to represent the domain (or the subpart of it) at
the right level of abstraction. In this particular case (i.e. concept specialization), the
main objective is to bring more precision in the description of the domain by
introducing new concepts. In our domain of study, to know the Web, this has been the
case for the concepts language, programming languages, markup language and
metadata system in 1998. Indeed, they have been transformed in a more specific
concept: XML the year after. This modification that brought more precision in the call
for papers has have an important impact on the submitted papers since 23 papers
dealing with XML have been accepted in 1999. However, this rapidity in the change
(only one year) can be explained by the analysis of the content of the papers
submitted in 1998. We first observe that the XML word appears mostly in the paper of
the track corresponding to the concepts that have changed (language, programming
language …). Concretely, there are 162 occurrences of the term XML in papers
related to programming languages, metadata systems and markup languages for a
total of 205 occurrences of XML in all the accepted papers. Furthermore, the study of
the abstract of these papers has highlighted that the concept XML was directly linked
to the concepts metadata, languages and markup languages through a subsumption
relation and terms metadata, markup language, and programming languages refer in
most of the case to XML in the content of the papers. This phenomenon combined
with the relevance of the XML language at this period of time has probably egg
WWW 1999 chairman to adapt the call for papers. This observation underlines the
relations between the interpretation of the domain (i.e. the content of the papers) and
the evolution of the domain itself (i.e. the call for papers). 7 concepts have been
specialized over the period of study and this evolution gave place to 16 new concepts
of the domain. Moreover, as it is the case for abstraction, the operation required more
or less time depending of the nature and the importance of the concept in the studied
domain.
Semantic Weight
Another important kind of evolution that has been highlighted by our study deals with
the notion of importance of the concepts in the domain. We call this phenomenon
concept emphasis. This property put the stress on the punctual tendency of the
evolution. In fact, at some time, concepts are more relevant for the domain than other
ones. Depending on the domain of interest, this “relevance” can be popularity,
profitability, technological improvements, etc. In our study, this is the case for
concepts like search, hypermedia or Semantic Web in 2002 but also ontologies
recently in 2006. This turns up at two different levels. First, it appears in the tracks of
the conference. In fact, since there have been so many accepted papers dealing with
these notions, two tracks were organized which underlines the importance or the
semantic weight assigned to these topics. Second, 83% of the accepted papers of the
other tracks contain at least one occurrence of the involved word which is also an
indication concerning the important aspect of the concept ontology in the domain. We
ESOE, Busan - Korea, November 2007 25
believe that this notion is really important and we will give an illustration in Section
4.
Semantic Distance
A more meticulous observation of the evolution of the domain of the Web through the
calls for paper of WWW’s events has permitted to emphasize the notion of semantic
distance between the concepts of the ontology. However, the distance we highlighted
is different from those proposed by Hirst-St-Onge [9], Jiang-Conrath [10] or Resnik
[14]. Actually, these metrics measure the distance between concepts that are linked by
at least a path composed by more than one arc in the graph of an ontology and the
objective is to estimate the closeness given their localization in the graph and the
number of arcs that separate them in this ontology. Nevertheless, we found, through
our empirical study, that the distance between concepts directly linked by the same
arc in the graph of the ontology varies. Actually, some concepts seem to be “closer”
(from the semantic point of view) than other ones although they are linked by the
same relation in the ontology. This turns up in the use of the words denoting these
concepts in the documents of the corpus. For instance concepts browser and
application appear more frequently in the papers than concepts tool and application in
1999 and both couple of concepts are bounded by the same relation (in this case the
relation of subsumption). Nevertheless, adequate metrics (different from those cited in
this subsection) are needed to catch this notion of semantic distance. For the time
being, we decide to consider words frequency. It means that we measure how many
times two concepts of a relation are cited together in the same kind of documents (i.e.
documents published the same year) and in the same context. Moreover, this distance
plays a key part to explain for instance the removal of concepts from the ontology. In
fact, concepts which are not relevant anymore for the domain, are getting further and
further form other concepts of the domain (i.e. the semantic distance is increasing).
Therefore, when a predefined threshold is reached, concepts are removed from the
domain. In the contrary, when concepts are very close, they can be replaced by a more
abstracted or specific concept if another appropriate threshold is reached.
Resistance
This other kind of evolution, called resistance to change, is a bit different from the
other characteristics presented so far. Actually, it has the particularity to be opposed
to evolution. This appears in our study in the ontology of 1998 and 1999. It reflects
also in the documents of the corpus. Indeed, there are 49 occurrences of the words
security in the papers accepted in 1999 which is very few. Furthermore, one paper
contains 26 occurrences of that word. This reveals that the notion of security was not
of utmost importance in 1998. Thus, following the natural aspect of the evolution
process, this concept should have been removed from the ontology representing
WWW 1999’s call for papers which is not the case as the concept security remained
in the call for papers in 1999. This proves that the chairman of WWW 1999 has
considered this notion as important for the field. The resistance to changes is also
present in other field mainly knowledge management [4], [12]. Nevertheless, the
resistance seems to vary according the concepts involved. Each concept resists
differently to evolution. The “coefficient” of resistance to change affected to each
26 International Workshop on Emergent Semantics and Ontology Evolution
concept is different. This introduces a notion of degree of freedom in the evolution of
the ontology. In fact, using this property, one can partially control the evolution of the
ontology. Thus, this newly introduced metrics should be determined rigorously by
domain experts. Furthermore, this phenomenon turns up under various forms in every
day’s life. For instance, for approximately 80% of the population, whales are seen as
fish and for only 20% of the people whales are mammals. In consequence, if we
follow the natural evolution process, in a significant period of time, all the people
should classified whales under fishes. Nevertheless, among the 20% of the people are
biologists (i.e. domain experts) that will permanently reject this evolution. This
proves the existence of such resistance to change and should be taken into account in
the ontology representing the domain mainly using adapted coefficient as shown in
section 3.2.
Speed of Change
The evolution of the domain takes place at different speeds. Some changes are rapid;
others are very slow and required several years. For instance, the specialization
concepts metadata systems, programming languages, and markup languages into
XML (as presented before) has taken only one year in the contrary, concepts browser
and tools have been abstracted into application in 2 years. We believe that the speed
of change is function of the coefficient of resistance to changes presented before. In
fact, if the coefficient is high, it means that the ontology should not change (or change
very little) which ensures a kind of stability in the ontology. However, if the same
coefficient is low, it will allow more flexibility in the evolution of the ontology. The
speed of change depends also on external factors like technology improvements. This
was the case for the concept Semantic Web; only one year after its definition it
became a key concept of the domain.
3.2 Modeling Features for Ontology Evolution
The various kinds of evolution we have highlighted through our empirical study, have
led us to the proposition of modeling primitives for ontology evolution. In this section
we describe these various modeling elements.
The proposed features can be classified into two different sets. Actually, we have
primitives that act on concepts (i.e. vertices of the ontology graph) and also primitives
that apply on relations (i.e. edges of that graph). The first set is made up of primitives
that put the stress on concepts emergence, concepts persistence and concepts
importance. So, first when a new concept emerge in a domain, it is important to know
the exact date at which the concept has appeared in the ontology. Second, concerning
persistence, two things are needed. On one hand, the emergence date and on the
other hand a duration determined manually by domain experts. The latter correspond
to a constraint of time the concept has to satisfy in order to be considered as
persistent. The last modeling feature that applies on concepts is related to concept
importance. We have decided to model this property using a coefficient called
importance. For the time being, this coefficient is computed based on the
occurrences and the repartition of the given concept in our corpus of documents. In
consequence, the more frequent its associated term is cited and the better the
ESOE, Busan - Korea, November 2007 27
repartition of this term in the corpus is, the higher the coefficient of importance will
be. We decided to limit the coefficient between 0 and 1 (1 representing a very
important concept).
The second set is formed by modeling primitives affecting relations between
concepts. These are related to the semantic distance and the resistance to changes.
Both notions are represented using coefficients. The semantic distance between two
concepts measures the evolution of the joint use of these two concepts in the corpus of
documents. This coefficient can vary from 0 to infinite but a maximum distance is set
by domain experts and if the distance reaches this particular value, the relation
between the two concepts is removed (i.e. an edge of the ontology graph is removed).
Moreover, if one concept becomes isolated in the ontology (i.e. it is no more linked to
any other concepts) it can be removed from the ontology. Concerning the coefficient
of resistance to changes, it must be defined by domain experts. This coefficient takes
its values between 0 and 1 where 1 denotes a very strong coefficient which prevents
every relation that is affected to evolve.
As OWL is the de facto standard for designing ontologies, we decided to study
how to represent the modeling elements presented in this paper using the OWL
language. Due to its powerful expressivity, OWL offers enough characteristics to take
all the presented features into account. Table 1 hereafter presents the various
modeling features, their associated datatype and the ontology notions they are applied
to. Observe that the types we use are the same than those contained in XML schema
definition.
Table 1. Modeling Features Summary
Element Name Datatype Affect
Emergence date xsd:dateTime concept
Persistence xsd:duration concept
Importance xsd:float concept
Semantic Distance xsd:nonNegativeInteger relation
Resistance xsd:float relation
However, OWL metamodel 3 [11 p.83] should be enriched in order to integrate the
above modeling features as basic OWL primitives. Concerning all features that apply
on concepts, three attributes should be added to the class Class of the OWL
metamodel. One attribute for representing the emergence date of a concept in the
ontology, a second one to express the persistence duration of a concept and finally a
third one for the importance of a concept. Moreover, these attributes must have the
same type than their associated elements (see table 1).
The two modeling elements related to relations, can be integrated to the OWL
metamodel by adding two attributes to the class Property. A first non negative integer
concerning the semantic distance and a second float for expressing the notion of
resistance to changes are needed. Nevertheless, the expressivity of OWL makes it
possible to easily express properties concerning concepts of an ontology without
3
The OWL metamodel we refer to is the one described using UML by Klein in his PhD Thesis.
28 International Workshop on Emergent Semantics and Ontology Evolution
modifying the OWL metamodel mainly using OWL Datatype properties but for
elements related to OWL properties it would be more problematic.
Another way to proceed would consist in using annotation properties or datatype
properties via datatypes defined in accordance with XML Schema datatypes to
express our concepts at ontology level. Nevertheless, this would require the
expressivity of OWL Full.
4 An Application to Web Information Retrieval
In this section we describe a real contribution of adaptive ontologies in the context of
information retrieval. Our formerly mentioned O4 approach [6], [8] implements an
ontology-based query expansion technique in order to improve the results, in terms of
relevance, when searching the Web. Actually, the query expansion phase is made
according rigorous expansion rules defined by taking into account terms of the query,
the form of the initial query and the relations that link the concepts of the query in the
ontology. The ontological relations implemented in this approach are on one hand the
well-known equivalence and subsumption relations which are already implemented in
OWL and on the other hand part-of and opposition relations which have all been
formalized in first-order logic and added to the Web Ontology Language as primitives
[8]. A first basic rule consists, given a basic query constituted by only one keyword,
in adding all the equivalent concepts of this keyword in the ontology. Nevertheless,
the ontologies we implemented so far were not able to evolve over time and thus do
not reflect the knowledge evolution of the domain they model. In consequence, the
choice of the right terms to put in the query was not fine enough. Due to the
properties of the evolution features we have presented in this paper, and mainly the
semantic distance and the semantic weight assigned to concepts of the ontology, we
will be able to refine even more this choice by selecting concepts which weights are
the highest since they are considered as the most relevant concepts of the domain. The
results of such a search will be more relevant because the more relevant concepts of
the domain will be added to the query.
Assume to illustrate this argument that an initial query “Web” will be submitted to
a Web search engine. If, for instance, the ontology we use to perform query expansion
contains two equivalent concepts for Web that are “WWW” and “Internet” with a
semantic distance from “Web” of 1 and 10 respectively. The system will select the
term “WWW” to put in the query since it is closer to the initial term “Web” than
“Internet” is close to “Web”. So, the expanded query “Web WWW” will be
submitted. Such expansion is judicious if we compare the different search results
associated to both queries “Web WWW” and “Web Internet”. Actually, pages
returned when the query “Web Internet” is entered are older and probably out of date
than pages returned corresponding to the other query. That shows that the integration
of domain evolution at ontology level will improve Web information retrieval at least
by giving right up to date information. Another basic example consists in filtering the
returned pages using the emergence date and the persistence duration of concepts that
constitute the query. Assume that the query “modem Internet” is submitted to a Web
search engine. The system would be able to return pages dealing with modems that
ESOE, Busan - Korea, November 2007 29
were published from the emergence date of modem in the domain of the Internet and
for the persistence duration of the modem concept.
This is all the more true for approaches implementing ontologies for tagging or
indexing information. Since the vocabulary for indexing or tagging is extracted from
ontologies, it has to be selected rigorously. Moreover, tags are usually chosen by
taking their popularity or any other properties that is domain dependent into account.
However, this kind of information was not provided by static ontologies.
Nevertheless, we have proposed an approach that has the advantage to integrate such
properties directly at ontology level. Therefore, if the concepts presented in this paper
will be integrated directly in ontologies, they will have a huge impact on approaches
dealing with tagging or information indexing.
5 Related Work
In the field of ontology evolution, relevant work has been carried out but two main
different approaches stand out. The first one, inspired by the work done in the
database field, considers ontology versioning. This problem has mainly been tackled
by Michel Klein [11]. He compared ontology evolution with database schema
evolution. The framework he proposed contains a set of operators, on the form of an
ontology, useful for modifying another evolving ontology. Klein also proposes a
change specification language based on the ontology of change operations. Moreover,
Avery and Yearwood, through their extension of OWL called dOWL [1], have
proposed a set of primitives to improve ontology versioning by facilitating the design
of dynamic ontologies.
The second approach for ontology evolution deals with consistency during the
evolution process. To this end, Ljiljana Stojanovic proposed a general methodology
for managing ontology evolution [15]. The process can be divided in 6 different
phases occurring in a cyclic loop. It enables handling the required ontology changes;
ensures the consistency of the underlying ontology and all dependent artifacts;
supports the user to manage changes more easily; and offers advice to the user for
continual ontology reengineering. Recently, Peter Plessers [13] described another
framework for managing consistent changes in ontology. This is done through the
definition of a Change Definition Language and the notion of version log. The former
is a temporal logic based language that allows ontology engineers to formally define
changes whereas the latter stores for each concept ever defined in an ontology the
different versions it passes through during its life cycle.
Besides, another interesting work has been carried out by Giorgios Flouris [5]. It
consists in applying approaches related to belief change to the ontology evolution
problem. The set of modeling features we propose introduces a new dimension in
ontology mainly by the introduction of the Semantic Distance between concepts of
the ontology. Nevertheless, our approach is different from the two approaches
existing in the literature which are reviewed in this section to know ontology
versioning and ontology evolution management. In our approach, we represent the
knowledge related to domain evolution in an ontology and show how this knowledge
can be exploited in Information Retrieval. Moreover, these new properties will allow
30 International Workshop on Emergent Semantics and Ontology Evolution
first to understand the evolution and will make it possible to anticipate future
evolution. Nevertheless, the dynamic ontologies we obtain can support ontology
versioning and moreover, since we formalized our ontologies in OWL, techniques for
change management can be applied too.
6 Conclusion
In this paper we have presented a domain analysis over a significant period of time
leading to a set of ontologies corresponding to the same views of a same domain over
different periods. We analyzed this set of ontologies in order to define new modelling
elements dealing with ontology evolution. Moreover, we also illustrate the potential
contribution of our proposition through an example dealing with information retrieval.
We believe that the evolution features we have defined consist in an important step
towards automatic ontology evolution. This will be possible if we find a way to
analyze the corpus of documents automatically. Nevertheless, our approach needs to
be strengthened mainly through the proposition of good metrics that will be able to
characterize as faithfully as possible the status of knowledge in a corpus of documents
from an evolution point of view. Therefore, our future work will concern on one hand
the definition of such metrics and on the other hand, the proposition of a formal set of
operators able to, given a corpus of documents, update automatically the appropriate
elements of the ontology we have introduced in this paper.
References
1. Avery, J., Yearwood, J.: dOWL: A Dynamic Ontology Language. In: Proceedings of the
IADIS International Conference WWW/Internet 2003, Algarve, Portugal, IADIS (2003)
985-988
2. Bachimont, B., Isaac, A., Troncy, R.: Semantic Commitment for Designing Ontologies: A
Proposal. In: 13th International Conference on Knowledge Engineering and Knowledge
Management (EKAW'02). Volume LNAI 2473., Sigüenza, Spain (2002) 114-1213.
3. Berners-Lee, T., Hendler, J., Lassila, O.: The Semantic Web. Scientific American 284(5)
(2001) 34-43
4. Biscaccianti, A., Renard, P.: The Cooperative Contextual Change Model: a Systemic
Approach to Implement Change while Preserving Stability. Cahiers du CEREN 4 (2003) 1-
16
5. Flouris, G.: On Belief Change and Ontology Evolution. PhD thesis, University of Crete,
Heraklion (2006)
6. Guelfi, N., Pruski, C.: On the use of Ontologies for an Optimal Representation and
Exploration of the Web. Journal of Digital Information Management (JDIM) 4(3) (2006)
7. Guelfi, N., Pruski, C., Reynaud, C.: Towards the Adaptive Web using Metadata Evolution.
In Calero, C., Moraga, M.Á., Piattini, M., eds.: Handbook of research on Web information
systems quality. Idea Group Publishing (2007)
8. Guelfi, N., Pruski, C., Reynaud, C.: Les ontologies pour la recherche ciblée d'information
sur le web: une utilisation et extension d'owl pour l'expansion de requêtes. In: Proceedings
of the Ingenierie des Connaissances 2007 (IC07) french conference, Grenoble (July 2007)
ESOE, Busan - Korea, November 2007 31
9. Hirst, G., St-Onge, D.: Lexical Chains as Representation of Context for the Detection and
Correction Malapropisms. In Fellbaum, C., ed.: WordNet: An electronic lexical database
and some of its applications, Cambrige, MA, The MIT Press (1998) 305-332
10. Jiang, J., Conrath, D.: Semantic Similarity based on Corpus Statistics and Lexical
Taxonomy. In: Proceedings on International Conference on Research in Computational
Linguistics, Tapei, Taiwan: Academia Sinica (1997) 19-33
11. Klein, M.: Change Management for Distributed Ontologies. PhD thesis, Vrije Universiteit
Amsterdam (2004)
12. Maurer, R.: Beyond the Wall of Resistance: Unconventional strategies that build support
for change. Bard Press (1996)
13. Plessers, P.: An Approach to Web-based Ontology Evolution. PhD thesis, Vrije Universiteit
Brussel (2006)
14. Resnik, P.: Using Information Content to Evaluate Semantic Similarity in a Taxonomy. In:
IJCAI. (1995) 448-453
15. Stojanovic, L.: Methods and Tools for Ontology Evolution. PhD thesis, University of
Karlsruhe, Universität Karlsruhe (TH), Institut AIFB, D-76128 Karlsruhe (2004)
16. Ashburner, M., Ball, C.A., Blake, J.A., Botstein, D., Butler, H., Cherry, J.M., Davis, A.P.,
Dolinski, K., Dwight, S.S., Eppig, J.T., Harris, M.A., Hill, D.P., Issel-Tarver, L., Kasarskis,
A., Lewis, S., Matese, J.C., Richardson, J.E., Ringwald, M., Rubin, G.M., Sherlock, G.:
Gene ontology: tool for the unification of biology. The Gene Ontology consortium. Nat
Genet 25(1) (May 2000) 25-29
32 International Workshop on Emergent Semantics and Ontology Evolution