SEMANTICS BASED RECONCILIATON
FOR COLLABORATIVE ONTOLOGY EVOLUTION
Georgios M. Santipantakis and George A. Vouros
Dept. of Information and Communication Systems Eng, University of the Aegean, Karlovassi, 83200, Samos, Greece
Keywords: Ontology evolution, Collaborative ontology engineering, Reconciliation, Validity rules.
Abstract: The objective of this paper is to present the SR-COE (“Semantics-based Reconciliation in Collaborative
Ontology Development and Evolution”) system for supporting knowledge workers/engineers to the
synchronous and asynchronous collaborative ontology development and evolution: This system provides the
necessary generic infrastructure for enhancing the deployment of any ontology development tool in
distributed settings with concurrent workers, and for applying any collaborative ontology engineering
method effectively. SR-COE exploits the semantics of the modification actions performed from the
different, concurrent contributing parties so as to actively support them to reach mutual agreed ontologies.
1 INTRODUCTION
Exploitation of knowledge represented by means of
ontologies aims to facilitate the knowledge
management processes within and across
organizations. To obtain a global view of a domain,
often people need to express their perspectives, in a
collaborative manner. However, it is likely to
happen that different persons provide conflicting
specifications for ontology elements. It is important
to locate the implied conflicts and provide support
for these persons to resolve them towards reaching
an agreement. Aiming to actively support knowledge
workers and engineers to shape a commonly
accepted conceptualization of a domain, this paper
presents the “Semantics-based Reconciliation for
Collaborative Ontology Development and
Evolution” (SR-COE) system: SR-COE exploits the
semantics of the modification actions performed by
knowledge engineers and workers, who act
concurrently on a single ontology. The system
detects actions with conflicting effects and
reconciles them, proposing alternative and conflicts-
free ontology versions. Semantics are being captured
by means of constraints/dependencies between the
effects of modification actions.
2 RELATED WORK
AND MOTIVATION
Methodologies towards the collaborative
development of ontologies propose concrete
methodological steps for facilitating collaboration
and reaching consensus. A wide range of approaches
is available, featuring techniques for locking
ontologies’ versions (Farquhar et al, 1996), centrally
accessible shared spaces (Kotis, Vouros, 2006),
and/or require collaborative parties to provide
arguments on the developed versions via
participating in argumentation dialogues (Tempich
et al 2005).
Several tools have been presented in the
collaborative ontology development domain (Sure et
al, 2002), (Arprez et al, 2001), (Bozsak et al, 2002),
(Domingue, 1998), (Ceusters et al, 2001), (Hotho et
al, 2006), (Tummarello et al 2006), (Sunagawa et al.
2003), (Auer et al, 2006) (Tudorache et al, 2008).
However, to the best of our knowledge, there is not
any tool that allows collaborative parties to
simultaneously and jointly modify the same
ontology version without any restriction (as far as
the creation and access to any version is concerned),
actively supporting people to locate and reconcile
conflicting specifications and views.
To facilitate collaborators’ awareness on the
status of the evolving ontology, some of the
methodologies/tools focus on providing a central
repository. Different versions of the same ontology
can be stored in this repository, possibly escorted
153
M. Santipantakis G. and A. Vouros G. (2009).
SEMANTICS BASED RECONCILIATON FOR COLLABORATIVE ONTOLOGY EVOLUTION .
In Proceedings of the International Conference on Knowledge Engineering and Ontology Development, pages 153-158
DOI: 10.5220/0002290501530158
Copyright
c
SciTePress
with arguments concerning changes made. Locking
and shared-space methods have inherent limitations:
(a) They do not conserve against conflicts that may
appear between versions, (b) they confine real
collaboration, as each user modifying an ontology is
the only one connected to it, and (c) they do not
actively support collaborative parties to inspect all
possible conflicts and reconcile different
perspectives.
To a greater extent than current tools/environ-
ments for ontology development/evolution we
require a system that shall facilitate collaboration by:
Imposing the minimum possible restrictions on
the collaboration and multi-party deve-
lopment/evolution process: The evolution,
creation and access to different versions of
ontologies needs to be made seamlessly, for all
participants, without directly affecting each
other.
Actively supporting the detection of conflicts for
each ontology version, and facilitating the
reconciliation of conflicts by exploiting the
semantics of the actions performed.
Being methodology-independent, therefore
generically applicable.
Facilitating the deployment of ontology
engineering tools in distributed settings, so as to
support collaboration.
Being seamlessly combined with any
argumentation framework.
We need to emphasize that the implementation of
such a generic system, acting as an infrastructure for
collaborative ontology development and evolution,
will allow the transfer of ontology engineering tools
from their “classic” standalone mode, to their
deployment in distributed settings.
To address these issues we have implemented
SR-COE. It actively supports people to collaborate
while being either connected or disconnected: Each
user holds a replica of an ontology, which is being
developed/evolved by using any preferred tool.
Modifications performed by remote and concurrent
collaborators on their local replicas need to be
merged: As local replicas may diverge, possible
conflicts between them must be detected and
reconciled. As a result of this reconciliation, all
users are aware of (a) the modifications made by
others, (b) of the possible conflicts that arise, and (c)
of the possible, alternative conflicts-free versions of
the ontology. We have to point out that the system
has to be able to support any state of the art
methodology that accentuate the active participation
of knowledge workers in the ontology engineering
lifecycle, (e.g. HCOME and DILIGENT).
3 BACKGROUND KNOWLEDGE
3.1 Semantics based Reconciliation via
Telex
The work presented in (Shapiro et al, 2004)
introduced a formalism for maintaining consistency
in distributed systems for data sharing. Based on this
formalism, work done in Telex (Benmouffok et al.,
2009) enables collaborative and distributed
development and evolution of “documents”. A
document may be of any form and format,
containing formal or informal specifications. In the
context of our work, a document is an evolving
ontology.
An application is a “middleware” between the
user and Telex (Shapiro et al, 2004). Telex is
instantiated in each collaborating site. Users by
means of their applications can modify their local
replicas of documents, either being online or offline,
while Telex takes the hard responsibility to merge
replicas. To do this, applications need to update
Telex with fragments of actions and constraints.
Then, Telex exploits the semantics of modification
actions which are being captured by constraints.
Constraints are application-specific invariants that
parameterize Telex. The types of constraints are
presented in the paragraphs that follow.
Telex takes care of replication, consistency,
storage and access control; collecting, transmitting
and persisting actions; detecting conflicts and
computing high-quality schedules.
Telex in each site maintains an Action-Constraint
Graph (ACG), where nodes are actions, and arcs
constraints between them. The actions represent
operations that users apply to replicas of a document
via their applications, and arcs represent constraints
between these operations. Constraints enable the
computation of possible, alternative schedules. The
ACG is formed by the union of all operations
performed in all sites: When a local update occurs,
the system replicates every action to the other
collaborating sites.
Table 1 shows the set of available types of
constraints (Benmouffok et al., 2009). The constraint
NotAfter indicates that an ordering between the
actions A and B must be maintained, such that no
schedule that commits to the execution of both
actions executes B before A. The constraint
Enables is an implication between the
corresponding actions. If a schedule commits to
execute B, then it must also commit to execute A
without considering any ordering between the
actions.
KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development
154
Table 1: Constraint types.
In case that A is aborted in a schedule this constraint
forces B to be aborted as well. Combining these
primitive constraints, the
Atomic, Casual
Dependence and Antagonism constraints are
generated. The
Casual Dependence constraint
defines that action B depends on A, and a schedule
commits to it iff it commits to A. The
Antagonism
constraint identifies a conflict between actions A and
B. This constraint defines that when a schedule
commits to execute action A, it aborts B and vice-
versa.
A set of actions is said to be in conflict, if they
form a “” (
NotAfter) cycle. This means that no
sound schedule can commit to the execution of all
these actions. The
Antagonism constraint is a
special case of such as cycle between the actions A
and B.
As shown in Figure 1, Telex comprises the
Scheduler, the Agreement module, the Logger and
the Transmitter. Each detected conflict triggers the
generation of schedules from Scheduler, which
“marks” each action to be either “aborted” or
“committed” in each schedule. For instance, the
detection of a pair of antagonistic actions drives
Telex to construct two alternative schedules, where
each of the antagonistic actions is either aborted or
committed.
A schedule may be selected by the local
application and be provided to the agreement
module as a proposal. Because of the asynchronous
communication, proposals may be different. The
agreement module is responsible though to help sites
reach an agreement. It is possible that some
applications require the selection of proposals based
on application specific criteria and/or policies. Telex
provides the freedom to developers to implement the
algorithm they need. The basic algorithm relies on
voting, so that user preferences are to be taken into
account. It is important to be mentioned that each
Telex module is instantiated per each locally “open”
document. We refer the interested reader to (Shapiro
et al, 2004) for further details.
3.2 Ontology Evolution via Validity
Rules
Considering the problem of ontology evolution in
the face of modification actions, authors in
(Konstandinidis et al., 2008) devise a framework and
algorithm for determining the effects and side-
effects of any such action. Although the framework
proposed is general, authors focus in a fragment of
RDF/S. To abstract from the underlying language
and develop a uniform framework, RDF is mapped
to First-Order Logic (FOL) expressions. The
representation of ontological facts by means of First
Order Logic predicates is shown in table 2
(Konstandinidis et al., 2008).
Table 2: Ontological facts as First Order Logic predicates.
The semantics of specifications are captured via
validity rules, which introduce a validity model.
Validity rules are encoded in the form:
),()(
,...,1 iiini
vuQvuPu
GG
G
G
=
where
i
v,u
G
G
are tuples of variables, P and Q
i
are
conjunctions of relational atoms of the form
R(w
1
,...,w
n
), and equality atoms of the form (w
k
=w
m
),
where w
1
,...,w
n
, w
k
,w
m
, are variables or constants.
An example of a validity rule is the following:
(
)
(
)()
wx,InstCwz,Domainzy,x,PI _
This rule states that given an instance (x z y) of a
property z whose domain is w, then x must be an
instance of w. All validity rules are specified in
(Konstandinidis et al., 2008).
The FOL specifications are equipped with closed
semantics, i.e., CWA (closed world assumption).
This means that, for two formulas p, q, if pq, then
p¬q. Abusing notation, for two sets of ground facts
U, V, we will say that U implies V (UV ) to denote
that U p for all p in V. Any expression of the form
P(x
1
,…,x
k
) is called a positive ground fact where P is
a predicate of arity k and x
1
,…,x
k
are constant
symbols. Any expression of the form ¬P(x
1
,…,x
k
) is
called a negative ground fact iff P(x
1
,…,x
k
) is a
positive ground fact. L denotes the set of all well-
formed formulae that can be formed in this FOL. We
denote by L+ the set of positive ground facts, and L-
SEMANTICS BASED RECONCILIATON FOR COLLABORATIVE ONTOLOGY EVOLUTION
155
the set of negative ground facts. An evolving
ontology is a set K
L+. In simple words, an
ontology is any set of positive ground facts. An
update is any set of positive or negative ground
facts. The application of any modification action to
an evolving ontology should result in a set of effects.
Each effect is a ground fact.
By definition, ontologies have two properties: (a)
they are always consistent (in the purely logical
sense) and (b) they imply only the positive ground
facts that are already in the ontology. The above two
properties together with the CWA semantics, imply
that (a)
P(x)
K
K P(x)
K ¬P(x) and
(b) P(x)
K
K ¬P(x)
K P(x)
An application of these properties is that, applying a
modification action with effect ¬P(x) to the ontology
K corresponds to removing P(x) from K. On the
other hand, updating an ontology with positive
ground facts corresponds to adding facts in K.
With the aim to specify and compute conflicts
using validity rules, following (Konstandinidis et al.,
2008), we use component sets. The component set
that corresponds to a validity rule c with respect to
some tuple of constants
x
G
is defined as
}}:,1|),(
...),(),({}0|)({{),(
21
constantznizxQ
zxQzxQkjxPxcComp
im
iij
<¬=
G
G
GGG
For example, for the rule
()
(
)()
wx,InstCwz,Domainzy,x,PI _
the derived component set would be
{{¬PI(x,y,z)}, {¬Domain(z,w)},{C_Inst(x,w)}}
As proposed in (Konstandinidis G. et al., 2008), a
ground fact
P(x) which is added in an ontology K,
would violate a rule c, iff there is some set V and
tuple of constants
u
G
for which ¬P(x)V and
V
()
uc,Comp
G
and for all V'
()
uc,Comp
G
, VV', it
holds that
K V'.
Let us for instance consider the following
scenario that exemplifies the above and reveals our
considerations for collaborative ontology evolution:
Let us consider George who performs an action that
adds the fact
PI(x,y,z) in his ontology replica, and
Peggy who performs an action with effect
Domain(z,w), in her own replica. Then, none of
these replicas violate the validity rule
() ()
(
)
wx,InstCwz,Domainzy,x,PI _
,
according to the above mentioned definition (given
that
C_Inst(x,w) does not hold in any of the
replicas). This is indeed so, since, given the
component set of this rule, the fact ¬
Domain(z,w)
(respectively, ¬PI(x,y,z)) validates the rule for
George (respectively Peggy). However, when the
two replicas are combined, the rule is violated, given
that in the merged version
PI(x,y,z) and Domain(z,w)
hold but
C_Inst(x,w) does not hold. Notice that if
C_Inst(x,w) holds, then both local views are
consistent, as well as the global view. However, it is
obvious that users cannot be “forced” to insert the
fact
C_Inst(x,w) in their replicas to assure the
validity of the merged version. In the merged
version this must be done to maintain consistency, or
else, alternative schedules must be constructed
where either
PI(x,y,z) or Domain(z,w) is aborted.
4 COLLABORATIVE
ONTOLOGY EVOLUTION
In order to apply the semantic-based reconciliation
method to the domain of ontology evolution,
possible modification actions and their constraints
must be defined.
Table 3 shows the constraints that derive from
the validity rules. The first column provides a
reference name, the second one contains the
component sets derived from the validity rules, and
the third one provides the corresponding constraints.
At this point it is important to notice two things:
(a) Constraints concern the effects of
modification actions, rather than the actions
themselves: Therefore, for SR-COE, nodes in ACG
are labelled with predicates from Table 2 and
conjunctions of them.
Subsequently, when we refer
to actions we actually refer to their effects
. Given a
schedule as described above, provided by Telex, we
can create a new ontology version from the schedule
itself. The generated version incorporates the effects
of actions performed by the different collaborating
parties and it provides a “landmark” for ontology
developers to proceed towards an agreed
conceptualization. This landmark is called a
checkpoint”. Checkpoints give the ability to the
users to “roll back”, i.e. to undo modifications.
(b) Some of the constraints in Table 3 have a
compound side, something which is not inherently
supported by Telex. For this purpose we introduce
the notion of the
compound (complex) action. A
compound action is formed by a conjunction of
actions. Its status, as it is determined by the Telex
scheduler, depends on the status of the actions it
consists of: If any atomic action contained in some
compound action is aborted, the compound is
aborted as well. A compound action is also
represented as an action in the ACG, along with the
constraints that bound it to the atomic actions it
consists of. For instance, given a compound action
x_and_y”, the constraints
x_and_y""x""
and
KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development
156
x_and_y""y""
are added to ACG.
Table 3: Derived constraints from component sets.
Considering again our example, the constraint
corresponding to the “Property instance and
Domain” rule will not be violated in any of the
replicas where
PI(x,y,z) or Domain(z,w) hold, but a
conflict will be detected in the merged version,
where both
PI(x,y,z) and Domain(z,w) hold and
C_Inst(x,w) does not hold.
Each of the collaborators modifies a local replica
of an ontology: The effects of modifications label
the nodes in ACG. Telex replicates and schedules
these effects to all sites. Any user may create a
checkpoint from a generated schedule and continue
evolving it. A checkpoint is published to remote
users as a new ontology version, by the time the user
who created it, starts modifying it. The set of shared
checkpoints defines the
versioning tree: Users may
select to evolve any checkpoint created by any of the
collaborating parties, creating a new set of
checkpoints, and so on. The depth and breadth of
this tree is subject to the rules and policies that the
collaborating community sets.
5 SR-COE ARCHITECTURE
An overview of the SR-COE architecture is
presented in figure 1.
Assuming that the user modifies a locally stored
ontology version, then, the following events occur:
The ontology parser detects new modifications
(the
diff) and calls the validity checker. These
effects will not directly affect other replicas.
Doing so, it allows users to modify the same
ontology with no restriction, concurrently.
The validity checker generates a new fragment
that is sent to Telex.
Figure 1: The overall architecture of SR-COE.
Telex takes control, to handle new fragments
appropriately. Modifications that do not raise
conflicts in schedules are appended to these.
Otherwise, alternative schedules are provided, if
possible.
The ontology updater module informs the user
for any new schedule and updates the locally
stored ontology model, on user’s request.
The paragraphs that follow present each SR-COE
module in detail.
The Ontology Parser module is responsible to
monitor the selected ontology version and retrieve
the effects of modification actions performed. When
a change occurs, the module identifies the elements
added or removed and provides them to the validity
checker module. This module is also responsible to
creating a checkpoint upon user’s request.
Given the new facts retrieved by the ontology parser,
the validity checker module is triggered to infer new
facts (i.e. inherited properties and transitivity of
subsumption relations), to produce constraints and to
validate the changes in the local replica.
In case that removed elements are detected, the
validity checker constructs antagonism constraints
between these facts and their negations. Constraints
generated by the validity rules are initially exploited
locally, to check for local conflicts. If no conflicts
are found, the validity checker proceeds to the
generation of a fragment containing new facts and
corresponding constraints to be sent to Telex.
Otherwise, the validity checker generates
antagonism constraints between the conflicting facts,
driving the computation of alternative schedules.
Doing so, users can continue modifying the locally
stored file, even if a conflict has occurred.
The Scheduler and Agreement modules of Telex
provide global consistency (given sound
constraints), and provide assistance towards
reaching an agreed conceptualization (Benmouffok
SEMANTICS BASED RECONCILIATON FOR COLLABORATIVE ONTOLOGY EVOLUTION
157
et al., 2009). Each user “runs” an instance of SR-
COE, and thus of Telex, that operates locally. Telex
is responsible to maintain the ACG, given the
fragments generated from validity checker module.
When the graph is updated by new nodes and the
corresponding constraints, the scheduler computes
new schedules, which are provided to the ontology
updater module.
The ontology updater connects the GUI of the
corresponding ontology engineering tool and SR-
COE. In our prototype implementation, it provides
information on the available schedules, as well as
information about modification actions (issuer,
action, effects) performed in the concurrent
collaborating sites. It notifies users about conflicts
detected and schedules built, without requiring users
to take any particular immediate action. However,
users may request to view a schedule and create a
new checkpoint which is added to the versioning
tree. Subsequently, the user can access this new file
via SR-COE, making this version available to the
rest of the community. Users proceed by choosing
the version to modify, creating new checkpoints, and
so on, until they reach an agreed conceptualization.
It is important to be mentioned that ontologies
comprise only positive facts that correspond to
committed actions in a schedule. Negative facts are
incorporated as nodes in ACG so as conflicts
between replicas to be detected.
6 CONCLUSIONS
To a greater extent than current tools/environments
for ontology development we have built a system
that facilitates collaborative ontology development
and evolution by: (a) Imposing the minimum
possible restrictions on the collaboration and multi-
party development/evolution process, allowing the
creation of different versions of ontologies and the
seamless access to these versions. (b) Actively
supporting the detection and reconciliation of
conflicts, by exploiting the semantics of the actions
performed. (c) Being methodology-independent,
therefore generically applicable. (d) Facilitating the
deployment of current ontology engineering tools in
distributed settings, supporting collaboration.
As described above, the validity rules presented
support the development and evolution of
lightweight ontologies using a rather simple model.
However, SR-COE can be easily customised by
incorporating any set of validity rules, to deal with
more expressive ontology languages.
An important step to be made concerns the study
of using SR-COE in conjunction with different
ontology engineering tools, maybe in the context of
different methodologies.
REFERENCES
Farquhar, A. et al, 1997. The Ontolingua Server: A tool
for collaborative ontology construction. Intl. J.
Human-Computer Studies 46, pp. 707–728 .
Kotis, K., Vouros, G., 2006. Human-centered ontology
engineering: The HCOME methodology, In KAIS, 10,
pp. 109-131.
Tempich, C. et al, 2005. An Argumentation Ontology for
Distributed, Loosely-controlled and evolvInG
Engineering processes of oNTologies (DILIGENT).
ESWC 2005, LNCS 3532, pp. 241–256 .
Sure, Y., et al 2002. OntoEdit: Collaborative Ontology
Development for the Semantic Web. ISWC 2002,
LNCS 2342, pp. 221–235 .
Arprez, J.C., at al, 2001. WebODE: a scalable workbench
for ontological engineering. K-CAP’01 .
Bozsak, E., et al, 2002. KAON - Towards a Large Scale
Semantic Web. 3rd Intl. Conf. on e-Commerce and
Web Technologies, LNCS 2455, pp. 304–313.
Domingue, J., 1998. Tadzebao and WebOnto: Discussing,
Browsing, and Editing Ontologies on the Web.
KAW’98, Banff, Canada.
Ceusters, W., et al, 2001. LinkFactory: an Advanced
Formal Ontology Management System. KCAP-2001,
Victoria, Canada.
Tudorache, T., et al, 2008. Collaborative Protege:
Enabling Community-based Authoring of Ontologies,
In Proc. Of the Workshop on Social and Collaborative
Construction of Structured Knowledge, ISWC 2008,
Banff, Canada.
Seidenberg, J., Rector, A., 2007. A Methodology for
Asynchronous Multi-User Editing of Semantic Web
Ontologies. K-CAP'07, pp. 127–134.
Hotho, A., et al, G., 2006. BibSonomy: A social bookmark
and publication sharing system. Conceptual Structures
Tool Interoperability Workshop at the 14th Int. Conf.
on Conceptual Structures, pp. 87-102.
Tummarello, G., et al., 2006. Enabling semantic web
communities with DBin: an overview. ISWC 06,
LNCS 4273, pp. 943–950.
Sunagawa, E., et al, 2003. An environment for distributed
ontology development based on dependency
management. ISWC’03, LNCS 2870, pp. 453-468.
Auer, S., et al, 2006. OntoWiki a tool for social, semantic
collaboration. ISWC, LNCS 4273, pp. 736-749.
Shapiro, M., et al, 2004. A constraint-based formalism for
consistency in Replicated Systems. OPODIS 04,
LNCS 3544, pp. 331-345.
Benmouffok, L., et al, 2009. Telex: A Platform for
Decentralised Sharing. In EuroSys 2009.
Konstandinidis G., et al, 2008. A Formal Approach for
RDF/S Ontology Evolution. ECAI-08, pp. 70-74.
KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development
158