DOMAIN ONTOLOGY EVOLUTION BY VERSIONING
Guilaine Talens and Danielle Boulanger
MODEME, Université Jean Moulin 6 cours Albert Thomas – BP 8242 69355 Lyon cedex 08, France
Keywords: Ontology creation, Ontology evolution, Ontology versioning, Semantic interoperability, Heterogeneous
information sources.
Abstract: To solve the semantic conflicts in the cooperation of heterogenous databases, we propose a multi-agent
system which contains domain ontologies. When a local source is added by an expert, ontology is
dynamically created by the system. The expert can also completed it. It evolves with the update of the local
base but also when an user performs a query. A process examines the query and creates temporary semantic
links. These must be validated or not by the user. The result validation implies the version creation from
temporary links. The modification and deletion of database elements by the expert perform the ontology
evolution. Our research work treats evolution by the version concept.
1 INTRODUCTION
The cooperation of heterogeneous information
systems requires advanced architectures able to
solve conflicts coming from data heterogeneity
(Sheth, 1999). We propose a Multi-Agent System
(MAS) to resolve semantic conflicts relatively to
evolutive domain ontologies following databases
evolution according to the dialogue between agents,
taking care of scalability issues. These interaction
protocols allowing ontologies evolution.
In our research, we focus on ontology versioning
to treat ontology evolution. Ontology versioning is
defined as ‘the ability to manage ontology changes
and their effects by creating and maintaining
different variants of the ontology’ in (Klein & al.,
2002). This functionality is important in scenarios
where users access an ontology in a distributed
manner.
In different previous research projects,
ontologies often evolve and the old versions are lost
because only the latest exists. Sometimes old and
new versions are archived but no mechanisms are
exhibited to expose the differences between
versions. Currently most of the works in the
ontology evolution and versioning captured their
ideas from schema versioning and schema evolution
in databases. But the content and the management of
ontologies are more difficult than database schemas
(Noy and Klein, 2003): ontologies incorporate much
more semantics and thus help to solve integration
problems between several information sources.
Since a few years, the use of ontologies to extract
implicit knowledge is a research-intensive approach
to overcome semantic heterogeneity difficulties in
the context of cooperation of heterogeneous
information sources. So, the characteristics of
ontologies do that the concept of evolution and
versioning in database schemas cannot be applied
directly in ontologies.
Ontology versioning has to focus on several
aspects :
- providing frameworks for ontology versioning
and evolution in the multi-ontology management
context,
- modelling and representing the details of
changes between ontology versions,
- specifying ontology changes operations and
analysing the impacts of these operations in different
contexts,
- developing algorithms and tools to compare the
different versions of a same ontology.
The current research works only treat few or
none of these points.
In first, we have briefly presented the problem.
In a second point, the domain ontology concept is
described. The third point focuses on ontology
evolution during the multi-database query execution.
The fourth point exhibits the evolution database
through version concept. Finally, we conclude by
highlighting the specific features of our contribution.
185
Talens G. and Boulanger D. (2009).
DOMAIN ONTOLOGY EVOLUTION BY VERSIONING.
In Proceedings of the International Conference on Agents and Artificial Intelligence, pages 185-190
DOI: 10.5220/0001557001850190
Copyright
c
 SciTePress

2 DOMAIN ONTOLOGIES
The ACSIS (Architecture Cooperative of Secure
Information System) architecture (Boulanger and
Dubois, 1998) allows the structural and semantic
conflicts resolution throws a multi agent system.
In our approach, the ontology of each agent
contains Data Descriptive Objects (DDO) and links
between these objects. An ontology can be defined
as a specific vocabulary and relationships used to
described some aspects of reality and a set of
explicit assumptions regarding the intended meaning
of the words vocabulary (Gruber, 1993) (Guarino,
1997). Recently, the Observer project (Mena and
Illarramendi, 2001) has described an ontology as the
concepts and the links which exist for an agent or for
several agents.
The DDOs contain the description of data from
local information sources as well as the access
primitives to this data. Local information entities
(relation, relation field, primary key, object type,
object attribute…) are described so that each
information source participating in the cooperation
process is represented by a set of DDOs.
The DDOs describe a class (in the object
oriented databases) or a relation (in the relational
databases).
The attribute DDOs include object attributes (it
could be object attribute or reference object attribute
that stores a pointer on an object) and relation
attributes (it could be primary key, foreign key or
relation attribute).
The links connect DDOs, according to
schematic, structural or semantic characteristics.
Three agent types interact:
- The Wrapper Agent (WA) ensures the
participation of local data to the cooperative
processes. Each WA is linked to a domain from a
local database. The DDOs and intra-base links form
its ontology.
- The Information Agent (IA) structures the
exchange between WAs during the processing of
global queries and semantic conflict resolution. Its
ontology is composed of the semantics links at the
global level (inter-bases links). Each IA groups WAs
according to semantic characteristics. An IA
accesses to at least one, and potentially many
information sources, and is able to collate and
manipulate information obtained from these sources
in order to answer the users and other IAs (Klusch,
2001). Each IA is a multi-domain agent. Its ontology
is formed by inter-bases links.
- The Interface Agent insures intermediation
between the user (expert or user role) and the
other agents:
* the User Agent manages the query, validates
the results and asks the re-execution of the query if
the results are not satisfying.
* the domain Expert Agent defines some intra-
base links, chooses the database type
(relational/object) and gives a representative name
of the domain.
Agents exchange information (see figure 1) by
the mean of interaction protocols to solve semantic
conflicts and to manage the evolution of domain
ontologies.
Database n°1
Database n°2
SGBD Access
Database n°3
SGBD Poët
Descriptive
level (database n°1)
Descriptive
level (database n°2)
Descriptive
level (database n°3)
Wrapper
User
Agent
queries
IA1
IA2
Database n°n
WA1
WA2
WA3
Descriptive
level (database n°n)
IAn
WAn
Expert
Agent
Database n°1
Database n°2
SGBD Access
Database n°3
SGBD Poët
Descriptive
level (database n°1)
Descriptive
level (database n°1)
Descriptive
level (database n°2)
Descriptive
level (database n°2)
Descriptive
level (database n°3)
Descriptive
level (database n°3)
Wrapper
User
Agent
queries
IA1
IA2
Database n°n
WA1
WA2
WA3
Descriptive
level (database n°n)
Descriptive
level (database n°n)
IAn
WAn
Expert
Agent
Expert
Agent
Figure 1: The different agents.
We explain our ontology through an example
created from a relational database (see figure 2). A
WA is created through the Expert Agent request to
add a new database. This new WA is named
“business” by the expert. A second WA named
“recreation” from an object oriented database (see
figure 3). These two WAs are attached to the same
IA.
Figure 2: Ontology created from relational database.
The DDOs and the schematic links are
automatically extracted from the database. The
dependence links allow the attribute DDOs
connecting to a relation DDO. The reference links
allow connecting the key DDOs. For example in the
figure 2, the id employee key of the work relation
references the id employee key of the employee
relation.
: relation DDO
Name : project
: relation DDO
Name : work
: relation DDO
Name : employee
: relation
Attribut DDO
Name : type
Type : String
: primary key
DDO
Name : name
Type : String
: primay key
DDO
Name : name
Type : String
:primary key
DDO
Name : id employee
Type : String
: relation
attribut DDO
Name : name
Type : string
: relation
attribut DDO
Name : salary
Type : real
Unit : Dollar
Dependence link
Referencement link
: primary key
DDO
Name : id employee
Type : String
: relation DDO
Name : project
: relation DDO
Name : work
: relation DDO
Name : employee
: relation
Attribut DDO
Name : type
Type : String
: primary key
DDO
Name : name
Type : String
: primay key
DDO
Name : name
Type : String
:primary key
DDO
Name : id employee
Type : String
:primary key
DDO
Name : id employee
Type : String
: relation
attribut DDO
Name : name
Type : string
: relation
attribut DDO
Name : salary
Type : real
Unit : Dollar
Dependence link
Referencement link
: primary key
DDO
Name : id employee
Type : String
: primary key
DDO
Name : id employee
Type : String
ICAART 2009 - International Conference on Agents and Artificial Intelligence
186

After the automatic extraction, the expert can
add semantic links:
Synonymy Links describe a similar sense
between two DDOs with different names (for
example between employee DDO and office worker
DDO).
Non Synonymy Links describe a different sense
between two DDOs with different names.
Similarity Links describe a similar sense between
two DDOs with the same name.
Homonymy Links describe a different sense
between two DDOs with the same name (name
DDO and name DDO if name is the attribute of
project and name is the attribute of employee).
Scale Links describe a same scale between DDO
with same name (wage DDO with Dollar unit and
wage DDO with Dollar unit).
Different Scale Links specify a unit existence
between two DDOs with same names (wage DDO
with Dollar unit and wage DDO with Dollar unit
when dollar is US or Canadian).
A link can have two states: permanent and
temporary. The state is temporary if the system
creates it; it is permanent, if it is created by a
database expert or if five users have validated the
system link creation.
Figure 3: Ontology created from object oriented database.
All these DDOs and links are created when a
new database is added: creation of a WA and the
WA attachment to an existing IA or creation of a
new IA (Talens, Boulanger, Séguran 2007).
These links can be also automatically created
during the query processing. A query example is
detailed to explain the ontology evolution.
3 ONTOLOGY EVOLUTION
3.1 Multi-base Queries
When a query is executed through the User agent,
different agents are contacted. Negotiation protocols
are defined for the communication between the
different agents: User Agent, Wrapper Agents and
interface Agents (Talens, Boulanger and Séguran,
2007). We only focus on the domain ontology.
The user writes through a User Agent the
following query: “SELECT * From employee
WHERE employee. salary=1500 euros AND
subscription . type=”year””. This query is sent to
Information Agents. The IAs exploit their ontology
in order to find interesting inter-bases semantic
links. After, each IA broadcasts to the network’s
Wrapper Agents the modified query (in the case of
relevant links have been identified, on the contrary
the initial query is sent).
Each WA extracts the different elements
(attribute, relation) of the query and compares them
to the DDOs. The intra-base synonymy and
homonymy links are also examined.
The business WA has the employee relation
DDO and the salary attribute DDO. A different scale
link is created (see figure 4) because the unit is
dollar in the attribute and euros in the query. A
virtual attribute is created to represent the different
scale link. A virtual DDO represents a query term; it
is only created for the representation of a temporary
link.
Figure 4: Different scale link between dollar and euros.
The subscription relation is not found. When one
or several elements are not found, a process is
applied for each of them to select a potential relevant
DDO. A temporary intra-base synonymy link is
created without the User Agent intervention thanks
to some schematic links (like the reference links and
dependence links) between the selected DDO and
the query element. This latter does not exist, so a
virtual DDO is created; it is a DDO only created for
the representation of this temporary link.
For each query element, if none DDO
corresponds in the ontology, there is a selection of
the refereed DDO thanks to dependence and
reference links, from detected DDOs. Moreover, if
there is no attribute specified in the query, or if the
specified attribute is equivalent to the attribute DDO
: class DDO
Name : person
: class DDO
Name : earner
: class DDO
Name : subscription
: class DDO
Name : subscription
: class
attribute DDO
Name : id
Type : String
: class
attribute DDO
Name : name
Type : String
: class
Attribute DDO
Name : function
Type : String
: reference attribute
DDO
Name : sportive
subscription
Type : String
: reference attribute
DDO
Name : sportive
subscription
Type : String
: reference attribute
DDO
Name : sportive
subscription
Type : String
: class
attribute DDO
Name : period
Type : string
: class attribute
DDO
Name : type
Type : string
: class attribute
DDO
Name : type
Type : string
Dependence link
Referencement link
: class attribute
DDO
Name : id subscription
Type : String
Is a
: different scale link
State : temporary
: virtual
attribute DDO
Name : salary
Type : real
Unit : euros
: relation
attribute DDO
Name : salary
Type : real
Unit : dollar
DOMAIN ONTOLOGY EVOLUTION BY VERSIONING
187

depending of the class/relation DDO, a temporary
intra-base synonymy link is created between the
reference DDO and a virtual class/relation DDO. If
the specified attribute in the query matches with the
attribute DDO depending of the reference DDO, a
temporary intra-base similarity link is created
between this attribute DDO and the attribute
element of the query.
In our example, thanks to the dependence link
from type attribute, the project relation DDO is
selected. A temporary synonymy link is created
between the project DDO and a virtual DDO named
subscription (query element). Another temporary
link is created; it is a similarity link between the two
type attributes. These two links and the different
scale link and the employee, salary, project and type
DDOs are returned to the IA.
For the second database (see figure 2), the
recreation WA has found the subscription class
DDO and the type attribute DDO. The employee and
salary DDO are not selected because from the
reference link the person and earner classes are been
potential relevant DDOs but they don’t contain the
type attribute. Only, the subscription and type DDOs
are sending to the IA.
When the WAs send some temporary intra-base
synonymy links to their IA, the IA could create
some temporary inter-bases synonymy links if the
same term exists in another WA of its acquaintance
network.
In our example, the recreation WA has the
subscription DDO and the business WA has the
project DDO (intra-base synonymy). A temporary
inter-bases synonymy link is created between these
two DDOs of these two WAs.
The IAs ask to the WAs to execute the query or a
part of it (selected DDOs of the previous step). For
the achievement, each WA (at the local level) and
each IA (at the global level) use intra-base or inter-
bases different scale links to translate data in the
expected format (if conversion functions are
detected in the DDOs).
In our example, the business WA performs the
following query: “SELECT * FROM employee
WHERE employee. salary=1500 AND project .
type=”year””. The recreation WA has only the query
concerning the subscription: SELECT * FROM
subscription WHERE subscription. type=”year””.
4 RESULTS
The User Agent organises the different answers of
each IA. The results are provided and also the
temporary links because they must be validated or
not by the user. The inter-bases and intra-base
synonymy links between project and subscription is
invalidated. The intra-base similarity link is also
invalidated. The different scale link is validated; the
user must insert a conversion function to convert
euros in dollar. Informations are sent to the
concerned IA and after to the WA. The modified
query is performed “SELECT * FROM employee
WHERE employee. salary=1500 euros” to the
business WA. For the other WA, the execution is
finished because the system does not require a
validation.
In the IA, a version of the synonymy link is
created. It contains the invalidated state, the time
and the identification of the user who has invalided
the synonymy link.
Figure 5: Synonymy link version.
In the WA, a similar version (see figure 5) is
created from intra-base synonymy link. A version is
also created from the similarity link between
subscription type and project type. A version is also
created from the different scale link. This version
contains the validated state, time, user and the
conversion function.
The system creates versions from the different
links because the new information is inserted in the
ontology only when it has been validated by several
users. In this way, the ontology consistency is
maintained.
For the synonymy link when it has been
invalidated by five users, it is transformed in a
permanent non synonymy link. In fact a non
synonymy link is created. If the temporary
synonymy link is validated five times, a new version
is created with the permanent state. When a link
version became permanent the WA domain ontology
evolves.
The advantage to store version is in a first time
the user validation or invalidation is not directly
inserted in the ontology. The second is when the
same query term is searched, the process to select a
DDO is not performed. The synonymy link is
directly proposed.
: synonymy link
State : temporary
: link version 1
State : invalidated
Time : 07/07/2008 15:39
User : user id 5
: virtual
attribute DDO
Name :
subscription
Type : String
: relation
attribute DDO
Name : project
Type : String
Version_of
ICAART 2009 - International Conference on Agents and Artificial Intelligence
188

5 LOCAL DATABASE
EVOLUTION
Different modifications can be performed on a
database. They are made by the local expert. He
manages the instance consistency because data
cannot be acceded. Only the database schema is
known by the system. The different modifications
relatively to a database are applied to the ontology of
the concerned WA and IA. The WA sends the
modifications from its IA, this one sends them at all
these WAs.
These modifications concern:
- Addition relation/class or attribute: adding a new
relation or class DDO and/or new attributes DDO.
The creation of similarity and different scale intra-
base links is processed as for the insertion of a new
database. The new DDOs are compared with the
other DDOs in order to create new links.
Figure 6: Versions from deleted DDO.
- Deletion relation/class or attribute: a version is
created from the concerned DDO with the "deleted"
state. A version is created from all the links
referencing this DDO. The state is validated or
invalidated, with mention of the time and by the
local expert. The state is the same as the previous,
the attribute deletion has no consequence on the link
state. The inter-bases links are not concerned
because they constitute global knowledge. To know
if a DDO is a synonym or an homonym of an other
DDO is very important for the IA because the query
is directly translated, none research is necessary.
For example, a synonymy link exists between
the name attribute and the family name virtual
attribute (query term). When the local expert deletes
the name attribute, a version is created from the
DDO with the deleted state (see figure 6). This
version creation brings about the version creation of
the entire links which referenced this DDO. In our
example, a creation version is made from the
synonymy link. The version creation is performed
for all the leaf versions of link. A version is only
created from the link version 1, its state is validated
because the previous version has the validated state.
- Modification:
- Attribute addition or deletion from a relation
or class (see the previous paragraphs),
- Class or attribute name: a version is
automatically created with the modified name state,
the new name, the time and the expert. A synonymy
link is also created between the DDO and the
version. For example, the local expert modifies the
name of the attribute salary. This modification
brings about the version creation from this attribute
and the creation of a synonymy link between the
attribute DDO and the attribute version. The state is
permanent because the modification has been
performed by the database local expert.
- Modification of attribute unit: Version creation
from the DDO with unit modification and version
creation from the scale and different scale intra-base
links. A DDO version is created from the attribute to
modify the attribute unit. The system automatically
creates successive versions from the different scale
links. The expert must give the conversion function
to be applied between the previous and new units.
- Modification of attribute type: a version is created
with the "type modified" state, the new type, time
and the database expert. A conversion function is
asked to the expert if scale or different scale links
exist for this attribute and therefore version creation
is performed to store the conversion.
6 RELATED WORKS
Several projects use ontologies and MASs. In
(Toinonen and Helin, 2003) an ontology describes
interactions protocols. Agents protocols dynamically
evolve with the ontologies. In our MAS system,
interactions protocols do not evolve but the
ontologies evolve.
The DASMAS project (Orgun et al., 2005)
provides a dialogue framework to treat semantic
interoperability in MASs. DASMAS runs with
several MASs. At each MAS corresponds a domain
ontology. The resolution conflict process uses a
negotiation protocol and WordNet Lexicon detects
semantical similar concepts in the heterogeneous
ontologies.
Some systems capture ontology evolution: Inside
the ontology or with ontology versioning. The
systems are also different in the manner to maintain
or not the changes between two successive versions.
SHOE (Heflin et Hendler, 2000) cannot keep track
of changes from a version to an other. SHOE
maintains each version of the ontology as a separate
: synonymy link
State : temporary
: link version 1
State : validated
Time : 07/07/2008 15:09
User : user id 5
: virtual
attribute DDO
Name : family
name
Type : String
: relation
attribute DDO
Name : name
Type : String
Version_of
: link version 2
State : validated
Time : 15/07/2008 16:09
User : local expert
Version_of
: DDO version 1
State : deleted
Time : 15/07/2008 16:09
User : local expert
Version_of
DOMAIN ONTOLOGY EVOLUTION BY VERSIONING
189

web page. The ontology designer copies the original
ontology file, assigns it a version number and adds
or removes elements as needed. On the contrary,
Protégé (Liang et al., 2005) keeps track and records
ontology changes within the ontology itself for
ongoing comparisons. OntoView (Klein et al., 2002)
system helps the user to manage changes in
ontologies and keeps ontology versions; it also
allows the users to specify conceptual relations
between different versions of concepts. The system
of Auer and Here (Auer and Here, 2006) keeps track
of different versions of an ontology and thus offers
merging operations or sub-hierarchies design.
7 CONCLUSIONS
Semantic conflict resolution with agents consists of
exploiting ontology links and DDOs. The set of
protocols permits the domain ontologies to evolve
by detecting new semantic links. The system
dynamicity comes from the detection of new links
during the query running and also from its ability to
follow local evolution databases. A prototype is
currently running on Jade platform (Bellifemine,
2002) and it permits to validate our proposition.
We apply the version concept in order to not
directly insert the modifications within the ontology.
The different versions constituting the historical
evolution are archived. The ontology versions keep
track of different modifications to better follow their
evolutions and to propagate modifications to
interconnected ontologies. The user’s modifications
become public after several validations from
different other users. The local databases evolution
is also managed by the version concept to store the
modifications and to propagate them along the all
existing semantic links.
REFERENCES
Auer S. and Herre H., “A Versioning and Evolution
Framework for RDF Knowledge Bases”, In
Proceedings of Ershov Memorial Conference, 2006.
Bellifemine, F.: Jade and beyonds. Presentation at
AgentCities Information Day 1, Lausanne, (2002)
Boulanger, D.,
Dubois, G., 1998. An Object Approach for
Information System Cooperation. Information
Systems, Vol. 23, n°6 (1998) 383-399.
Gruber T.R., 1993. A translation approach to portable
ontologies. Knowledge acquisition, vol. 5, n°2, , p.
199-220, 1993.
Guarino N., 1997. Understanding, Building, And Using
Ontologies. International Journal of Human-
Computer Studies, vol. 469, n°2-3, Feb./March, 1997.
Heflin J. and Hendler J., 2000. Dynamic Ontologies on the
Web. In Proceedings of the seventeenth national
conference on Artificial Intelligence, AAAI 2000,
Austin, 2000.
Klein, M, Noy, N., 2003. A Component-Based Framework
for Ontology Evolution. Technical Report, IR-504, CS
Dept, Vrije Universiteit, Amsterdam, 2003.
Klein M., Kiryakov A., Ognyanov D. and Fensel D., 2002.
Ontology versioning and change detection on the web.
In 13th International Conference on Knowledge
Engineering and Knowledge Management, EKAW02,
Spain, 2002.
Klusch, M., 2001. Intelligent Agent Technology for the
Internet : A Survey. Journal on Data and Knowledge
Engineering. Special Issue on Intelligent Information
Integration, D. Fensel (Editor), Vol. 36, n°3, Elsevier
Science (2001).
Liang Y. Alani H. and Shadbolt N.R., 2005. Ontology
change management in Protégé. Proceeding of
Advanced Knowledge Technologies Doctoral
Colloquim, United Kingdom, 2005.
Mena E., Illarramendi A., 2001. Ontology-Based Query
Processing for global Information Systems. Kluwer
Academic Publishers, June, (2001).
Noy N.F. and Klein M., 2003. Ontology evolution: Not the
same as Schema Evolution. Knowledge and
Information Systems, 5, 2003.
Orgun B., Dras M., Nayak A.,2005. DASMAS – Dialogue
based Automation of Semantic Interoperability in
Multi Agent Systems. Australian Ontology Workshop,
AOW’2005, Sydney, Australia, 2005.
Sheth A., 1999. Changing Focus on Interoperability in
Information Systems: from System, Syntax, Structure
to Semantics”, In Interoperating Geographic
Information Systems. M F. Goodchild, M.J.
Egenhofer, R. Fegeas, C.A. Koffman (editors), Kluwer
Academic Publishers, 1999.
Talens G., Boulanger D., Séguran M., 2007. Domain
Ontologies Evolutions to Solve Semantic Conflicts. In
Ontologies-Based Databases and Information Systems
, volume 4623/2007, LNCS Springer Berlin
/Heidelberg, ISBN 978-3-540-75474-9_4, September
(2007), pp 51-67.
Toivonen S. and Helin H., 2003. Representing Interaction
Protocols in DAML, In van Elst, L., Diagnum, V., and
Abecker, A. (Eds.): Agent-Mediated Knowledge
Management: Selected Papers from AAAI 2003
Spring Symposium, volume 2926 of Lecture Notes in
Artificial Intelligence, Berlin, Germany, 2003.
ICAART 2009 - International Conference on Agents and Artificial Intelligence
190