ONTOLOGIES AND COMMUNITIES CO-EVOLUTION
IN INFORMATION SYSTEMS
Francesca Arcelli Fontana, Ferrante Formato and Remo Pareschi
University of Milano Bicocca, Milan, Italy
University of Sannio, Benevento, Italy
University of Molise, Campobasso, Italy
Keywords: Communities, Complex networks, Ontologies, Knowledge management.
Abstract: Communities and ontologies are both concepts that have acquired strong momentum since the coming of
age of new media such as Internet and the Web. They have become more relevant in a situation where
growing communities and creating information categorizable through ontologies is made much easier and
faster compared to what was possible before. In spite of this concomitance, the roles they have played in
this information-rich environment have been so far not only different but also largely antithetic. The one
played by communities is dynamic, and views information as something which is constantly changed and
re-created by the agents that produce it. By contrast, the one played by ontologies views information in
terms of its management at the meta-level through categories and concepts hierarchies, and it assumes that
the ontology remains static, or changes very slowly as a consequence of decisions taken by the domain
experts that control it. Given that information change is generally community-driven and this brings the
clear necessity to make communities and ontologies interact. We propose to pursue this goal through a
knowledge management approach, where the interaction between communities and ontologies is
implemented as a knowledge life-cycle that leads to the creation of new concepts in the ontology as a
consequence of the evolution of the information spaces constantly extended and re-created by the
communities.
1 INTRODUCTION
Ontologies have since a long time provided a
powerful tool to organize knowledge. At a
philosophical level, ontology is the most
fundamental branch of metaphisics. It studies being
or existence and its basic categories and
relationships, to determine what entities and what
type of entities exist. At the more specific level of
knowledge representation and knowledge
management, ontologies identify concepts applied to
specific domains and organized as graphs via
relationship links. A typical example of an ontology
as shown in Figure 1 is given by an automotive
ontology, organizing concepts used by enterprises
operating in the automotive industry.
Domain ontologies are traditionally the product of
panels, teams and committees of domain experts and
knowledge engineers. As such they are designed,
maintained and evolved by these organized groups
on the basis of the needs and objectives of the larger
organizations they belong to. However, in a situation
where organizations and corporations act less and
less as the closed information sylos of the industrial
age and are indeed compelled to re-act and co-act
with an information-rich environment in order to
prosper and survive, this approach appears too rigid
and static. The desideratum would rather be one
where the communities that provide the
user/stakeholder bases for the products and services
of organizations give also the input for the evolution
of their conceptual infrastructure, so as to effectively
capture and reflect dynamically the evolution of user
needs and market trends. Domain experts and
knowledge engineers would still be involved, but in
an effort of combining and rationalizing knowledge
effectively emerged from the bottom, rather than of
imposing concepts more or less arbitrarily decided at
the top. Thus, the ultimate goal is to make perfectly
synchronous the alignment between organizations
and their user and stakeholder communities, and to
fully exploit the enormous potential for concepts
453
Arcelli Fontana F., Formato F. and Pareschi R..
ONTOLOGIES AND COMMUNITIES CO-EVOLUTION IN INFORMATION SYSTEMS .
DOI: 10.5220/0003106404530458
In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2010), pages 453-458
ISBN: 978-989-8425-29-4
Copyright
c
2010 SCITEPRESS (Science and Technology Publications, Lda.)
creation deriving from such information sources as
the new digital media, Web and the Internet in
primis.
Figure 1: AN automotive ontology.
As we aim to show here, achieving this goal has
some interesting consequences at a general
foundational level, since it implies reconciling the
“simple” and the “complex”, by making rational
design and planning interact with the turbinous
growth patterns of real life (even when in the form
of “digital” life). Indeed it turns out that this is
obtained by combining in a consistent way two
different scientific traditions: one, rooted in
philosophical logic and knowledge representation,
and concerned indeed with the “simple” and the
“rational” namely, the study of ontologies and of
their logics; the other, rooted in network theory,
focused on the “complex” and the “emergent”
namely, the study of communities and of social
networks.
The theoretical building blocks of our approach can
be described as follows:
1. An “object” level where communities create
content and information, and a “meta-level”
where content and information is classified
into ontologies;
2. A “knowledge life-cycle” that defines the
interaction between the two levels in 1. (that’s
where the main novelty of our approach lies)
and makes possible to up-raise the process of
information-creation at the object level into a
process of knowledge-creation at the meta-
level through the introduction of emergent
concepts concepts that, once certified and
stabilized by teams of experts, can flow back
into communities where they are adopted and
shared.
1.1 Related Works
The problem of ontology evolution has been
addressed by several authors in the literature. For
example in (Stojanovic et al., 2002), the author
claim that since generally ontologies grow in size,
this requires a well structured ontology evolution
process and they introduce the concept of an
evolution strategy encapsulating policy for evolution
with respect to user’s requirements.
In (Noy and Klein, 2004), the authors, in the
context of ontology-evolution frameworks, analyze
the similarities between database-schema evolution
and ontology evolution which allow to develop an
extensive research in schema evolution. In (Klein
and Noy, 2003), the authors address the importance
of ontology evolution in distributed development
and they present an ontology of change operations,
which is the kernel of a framework they proposed.
For what concerns with our research, in a
previous work (Arcelli et al-a, 2009) we define a
model by which ontologies evolve through Web
community extraction. While in another work
(Arcelli et al-b, 2009) we have introduced a
methodology based on complex network
parametrization, that studies the evolution of
complex networks through an operator on graphs,
whose purpose is to equalize meta-ontologies in the
model we have proposed. Here, we describe an
approach, with a wide scope both in terms of
foundations and applications, that is based on the
techniques and the apparatus we have described in
our previous works. Hence our contribution here
comes in the form of a research manifesto.
2 REPRESENTING
ONTOLOGIES AND
COMMUNITIES
Our view both of ontologies and of communities is
information-driven: they are identified with the
information they contain, either because they
produce it (in the case of communities) or because
they categorize it (in the case of ontologies).
Furthermore, both communities and ontologies can
be represented as networks (directed graphs). This
common formal representation makes it easy to
model the interaction between the two levels, yet it
does not hinder us from identifying specific
topological properties of the different types of
networks that will be used to represent, respectively,
ontologies and communities.
KEOD 2010 - International Conference on Knowledge Engineering and Ontology Development
454
Indeed, ontology networks are typically
characterized by a fixed number of nodes,
corresponding to concepts, connected via a uniform
distribution of links. This follows from their nature
of networks planned and designed in a controlled
fashion, with the aim of providing a complete and
consistent conceptualization of a certain piece of
knowledge.
By contrast, community networks are typically
characterized by such phenomena as preferential
attachments, meaning the fact that some nodes will
be pointed to more than others as a consequence of
their role of “hubs” and “leaders”. Furthermore,
community networks will grow dynamically as more
members join the community. As a very important
caveat, it should be made clear that the notion of
community that we adopt here not only assumes
networks as a form of representation, but is itself a
specialization of the notion of network: in fact, we
adhere to the view, coming from the tradition of
network theory, that a community can be defined in
topological terms as a region of a dynamic network
where links are denser than in the surrounding
regions. In other words, communities are directly
identified with highly interconnected regions of
dynamic networks, as shown in Figure 2. This
allows us to model as communities social networks
whose nodes map directly into human individuals,
such as family clans, but also digital communities
where the role of humans is crucial but indirect, in
that the primary community members are Web sites
pointing one to the other. As we shall discuss later
on, the most immediate applications for our
approach to co-evolution of communities and
ontologies are indeed in the domain of this kind of
Web communities.
From the formal standpoint of network theory,
the uniform link distribution of ontologies
corresponds to networks-as-lattices as utilized in
knowledge representation, see for instance the
classical book by (Sowa, 1999) for a general
overview of the subject.
Figure 2: Communities within a network.
The preferential attachment behavior of
community networks is formally accounted for by
the scale-free networks recently studied by Barabási
and his associates (Albert and Barabasi, 2002).
3 KNOWLWDGE LIFE CYCLE
Precisely because we view the world of
organizations, expert teams and ontologies on one
side, and the world of communities and emerging
concepts and experiences on the other, as
communicating rather than as separated, we aim to
define a model through which they can fruitfully
interact, thus making possible the circulation and re-
creation of conceptual knowledge. This healthy
circulation is the opposite of the corporate ailment of
knowledge stagnation, which takes place whenever
organizations lose touch with the needs and feelings
driving the communities of the stakeholders they
depend on perhaps the most deadly of corporate
disesases in an era where enterprises can effectively
compete on communication and knowledge transfer.
The model of lifecycle that we adopt is itself an
adaptation of the well-known “Double-loop
Learning” model developed by (Argyris and Schön,
1978), which has found vast and effective
application in the management of many types of
knowledge processes in a variety of organizations.
We specifically apply it to the interaction between
organizations and the social networks of
stakeholders existing at their borderline a
phenomenon which has emerged forcefully with the
extended communication spaces of the new
millenium.
Argyris and Schön distinguish between single-
loop and double-loop learning, related to Gregory
Bateson’s concepts (Bateson, 1979) of first and
second order learning. In single-loop learning,
individuals, groups, or organizations modify their
actions according to the difference between expected
and obtained outcomes. In double-loop learning, the
entities (individuals, groups or organization)
question the values, assumptions and policies that
led to the actions in the first place; if they are able to
view and modify those, then second-order or double-
loop learning has taken place. Double- loop learning
is the learning about single-loop learning.
ONTOLOGIES AND COMMUNITIES CO-EVOLUTION IN INFORMATION SYSTEMS
455
4 ONTOLOGY (CO)-EVOLUTION
Quite obviously, single-loop learning takes place in
applying an ontology to the domains it is supposed
to categorize. Take for instance a wine ontology and
consider a directory of Web sites related to wine,
including wine sellers, wineries, wine clubs etc.
Then one simple procedure to learn how to use the
ontology is as follows:
we select Web sites from the directory and
we associate them with nodes in the
ontology;
the concepts in ontology sites get “trained”
with the content in the Web sites, through
some classifier algorithm such as bayesian
inference, neural networks, or support vector
machines;
by following the links of the Web sites, we
apply the trained concepts to new content:
o if they classify according to expectation
then we have reached the appropriate
training of the ontology and thus we have
learned how to use it;
o Otherwise we might need to refine their
training so as to effectively make them
capable to clasify Web sites as expected.
Now, what about double-loop learning?
According to the definition, this must question the
structure of the ontology itself by bringing as a
consequence the introduction of altogether new
concepts. As a matter of fact, this is what happens
whenever teams of experts revise ontology
structures in order to adapt them to changes in
industry and market trends. Thus, such changes are
generally re-active to mutated conditions in the
environment; therefore, what is still missing is a
sound methodology to alert the experts of the need
of change and drive them in the right direction. In
order to answer to this need, we apply double-loop
learning as follows:
let’s assume that we have applied single-loop
learning as above. This means that we have
partitioned a part of the Web into “concept
graphs”, identified by the scope of
application of the concepts in the ontology;
then let us explore this portion of the Web
watching for communities (namely, highly
interconnected regions) and distinguishing
two cases:
o communities corresponding to existing
concept graphs;
o communities that do not fit with existing
concept graphs (even if they may be
partially overlapping with existing graphs)
if the second such case occurs, then this is an
indication that we are in front of one or more
new concepts, and that the overall current
architecture of the ontology must be revised;
it is then the work of the experts to acquire
this input and elaborate it through the various
techniques available, starting from the
inspection and the analysis of the Web sites
belonging to the uncovered communities, and
to extend and revise the ontology
accordingly.
Both these steps are summarized in Figure 3, which
depicts double-loop learning as applied to
ontology/community co-evolution.
Figure 3: A double-loop learning system for
community/ontology co-evolution.
As an example, suppose that we find a
community of Web sites which is not covered by
any existing concept in the corresponding wine
ontology. On the other hand, a sub-region of this
community is indeed covered by a concept graph
corresponding to the concept of “White Zinfandel”
(the rosè wine from California). The remaining part
of the community is characterized by content related
to Italian ham and salami. On the basis of further
content analysis, this may be taken as an indication
that White Zinfandel lovers see Italian ham and
salami as a suitable food match for their favourite
wine, and that creating a corresponding “menu
KEOD 2010 - International Conference on Knowledge Engineering and Ontology Development
456
concept” may be relevant and approparite, with
possible applications to the design of new products
packaging White Zinfandel and Italian ham and
salami to be distributed in wine shops, food stores
and shopping malls.
It should be pointed out that, beside double-loop
learning, this approach could be viewed as fitting
within other knowledge management methodologies
such as, in particular (Nonaka and Takeuchi, 1995)
Knowledge Spiral, which can be considered itself as
evolving further the concept of double-loop learning.
The Knowledge Spiral defines a cycle of four
phases, given by Knowledge Internalization,
Knowledge Socialization, Knowledge
Externalization, and Knowledge Combination. In
our context, Knowledge Internalization and
Knowledge Socialization play the role of single-loop
learning as viewed above, namely as learning to
apply the ontology to the relevant part of the Web,
while Knowledge Externalization and Knowledge
Combination play the role of double-loop learning in
the different phases of identification of a new
concept from the Web and of consequent re-design
of the ontology.
Finally, the strong use of tools from Information
Technology and Artificial Intelligence to support the
automation of the different phases of the learning
cycle, such as content classification and analysis
algorithms, suggests also, for the purpose of its
general support, the definition of a Knowledge
Management IT Architecture in the sense of
(Borghoff and Pareschi, 1998).
5 APPLICATIONS
The idea of leveraging in a systematic way the eco-
system that connects organizations with their
surrounding communities in order to pursue concept
creation has a very wide potential, with applications
that, in different ways, reach the very core issues of
innovative design of products and services. Here we
highlight briefly two specific domains that appear
as particularly relevant: user-driven innovation and a
“community-oriented” version of the Semantic Web
project.
User-driven Innovation. There are two main
approaches to product innovation. In the so-called
“linear model” the traditionally recognized source is
manufacturer innovation. This is where an agent
(person or business) innovates in order to sell the
innovation. Another source of innovation, only now
becoming widely recognized, is end-user innovation.
This is where an agent (person or company)
develops an innovation for their own (personal or in-
house) use because existing products do not meet
their needs. Eric von Hippel has identified end-user
innovation as, by far, the most important and critical
in his classic book on the subject, The Sources of
Innovation (von Hippel, 1988). One outstanding
example of end-user innovation is open-source and
free software.
However, while many users may correctly
identify the need of innovation, they may lack the
technical skills or the economical means or simply
the will to innovate. Ideally, this situation could
offer excellent opportunities for manufacturers to
innovate effectively, if the could listen carefully
enough to their user communities, thus providing an
intermediate model between manufacturer and end-
user innovation. Of course, this idea is not new but
so far it has not been obvious how to put it in
practice. User groups as supported and implemented
nowadays by many enterprises go in this direction,
but, again, they imply the willingness of users to
organize themselves in somewhat formal structures,
which may be less productive and creative with
respect to the totally free format given by
communities.
Double-loop learning to make innovative
concepts emerge from communities and enter, with
an effect of creative disruption, corporate ontologies
may provide an important basis to evolve this
potential for product innovation into a fully
practicable methodology.
Semantic Web. The Semantic Web http://www.
w3.org/2001/sw/ is a project, managed “from-the-
top” by standard committees and research
institutions, to make the Web fully
“understandable”. (For an overview, from the point
of view of the Semantic Web founders, of where the
Semantic Web stands since its inception in the very
early years of this millenium see (Berners-Lee et al.,
2006). In this way, software agents could inspect
content of the Web pages and automate e-business
and e-commerce actions. On the other hand, by
moving from the “bottomground of people and
communities, the primeval Web (so called Web 1.0)
has evolved on its own into something completely
different, Web 2.0 namely the Web of blogs,
social networks and personal spaces. There is a
general consensus that Web 2.0 is, first and
foremost, about people, and is neutral and open to
any kind of technologies or standards as long as they
provide support to people-oriented applications.
We view our approach as instrumental to
reconciling the quest of semantic clarity initiated by
ONTOLOGIES AND COMMUNITIES CO-EVOLUTION IN INFORMATION SYSTEMS
457
the Semantic Web project with the explosive growth
of people-oriented Web 2.0. The point is that the
Semantic Web, as originally conceived, implies a
strong management of the information available on
Web sites, by annotating it manually with semantic
meta-information such as XML tags, ontologies and
“resource-description frameworks”. This contrasts
with the way people use the Web, and Web 2.0 in
particular, that is essentially for communication and
personal networking, caring a lot about content and
not too much about meta-content. By providing a
way to co-evolve communities and ontologies, our
framework can be exploited to automate the creation
of a “meta-web” where the burden of semantic
annotations is taken away from the users.
6 CONCLUSIONS
As in all the research manifestos, the conclusion is
the beginning of… the beginning, and the main thing
that can be said is that we expect much from what
has to come. But, just to summarise a bit, we state
again the goal of this research program: which is of
viewing communities and ontologies, two concepts
that have both gained strong momentum through the
coming of age of the new media, as fully
complementary even if they move from apparently
distant premises, viz. emergent behavior in one case
and rational design and planning in the other. The
result shall be a novel knowledge lifecycle aimed at
avoiding knowledge stagnation through the constant
generation of fresh concepts, and the consequent re-
design of the ontologies that host them a result
obtained by combining in a non-intrusive way the
creative force of communities and the rational
design of knowledge teams.
REFERENCES
Albert, R., and Barabási, A., 2002. Statistical mechanics
of complex networks. Review Modern Physics, 74, 47.
Arcelli, F. Formato, F. and Pareschi, R., 2009a. Ontology
Engineering: Co-evolution of Complex Networks with
Ontologies, Proceedings of the Workshop on
Ontologies for e-Tchnology (OET 2009), Italy.
Arcelli, F. Formato, F. and Pareschi, R. 2009b. Equalizing
the structures of web communities in ontology
development tools. Proceedings of the International
Conference on Intelligent Systems Design and
Applications, (ISDA’09), Italy.
Argyris, C., and Schon, D., 1978. Organisational learning:
A theory of action perspective. Addison Wesley,
Reading, MA.
T. Berners Lee, N. Shadboldt and W. Hall, 2006. The
Semantic Web Revisited. IEEE Intelligent Systems, 21
(3).
Bateson, G., 1979. Mind and Nature: A Necessary Unity,
Advances in Systems Theory, Complexity, and the
Human Sciences. Hampton Press. ISBN 1-57273-434-
5.
Borghoff, U. M., and Pareschi, R., 1998. Information
Technology for Knowledge Management. Springer-
Verlag, Berlin and Heidelberg, Germany
von Hippel, E., 1988. The Sources of Innovation. Oxford
University Press, Oxford, UK.
Klein and Noy, 2003. A Component-Based Framework
For Ontology Evolution. Proceedings of the IJCAI
Workshop on Ontologies and Information Sharing,
Seattle, WA.
Nonaka, I., and Takeuchi, H. (1995) The Knowledge
Creating Company. Oxford University Press, Oxford,
UK.
Noy and Klein, 2004. Ontology Evolution: Not the Same
as Schema Evolution, Journal of Knowledge and
Information Systems, Vol.6, N.4, Springer London
Sowa, J. F., 1999. Knowledge Representation: Logical,
Philosophical, and Computational Foundations.
Brooks Cole Publishing Co., Pacific Grove, CA.
Stojanovic et al., 2002. User-Driven Ontology Evolution
Management, Knowledge engineering and knowledge
management: Ontologies and the semantic web, LNCS
2473/2002, Springer.
KEOD 2010 - International Conference on Knowledge Engineering and Ontology Development
458