Modeling a Complete Ontology for Adaptive Web Based Systems Using a

Top-Down Five Layer Framework

Bujar Raufi, Florie Ismaili, Xhemal Zenuni

South East European University, Ilindenska nn 1200 Tetovo, Macedonia
E-mail(s): [email protected], [email protected], [email protected]

Abstract. The web pages of today strive to user. The main source used to extract user
think out of the box by creating different look- behaviour is the web server logs. These
and-feel for each particular user. Most of them systems or approaches are known also as
still are a pure cast of HTML tags, however reactive systems or approaches [12].
some web applications tend to organize their - And finally, in Dynamic Web Based
structure and content based on user access Systems, where besides them being adaptive,
patterns. These systems are known as Adaptive there is no predefined data presentation, but
web based systems. the adaptation and page synthesis is being
In this paper we present a complete ontology perfomed out of elementary pieces of
(Concept Domain) for Adaptive Web Based information (atomic units) such as: Text,
Systems based on Top-Down five layer Multimedia, Links etc.
framework. We also test the inference in such The last type of adaptive web based system is
ontology and conclude that the findings fulfill of particular importance to our research as we are
the requirements of our proposed framework and going to see later when we present our
that this ontology is suitable for general purpose framework and ontology. The contribution of
Adaptive Web-Based Systems. this ontology is that it offers expressivity,
modularity and interoperability as required by
Keywords. Adaptive Web-Based Systems, our framework; as well as its suitability for
Ontology, Semantic Web general-purpose Adaptive Web-Based Systems.
The rest of this paper is organized as follows:
1. Introduction In section 2. we give a brief description of
related work on semantic web ontologies
Most of the web pages we see today represent regarding adaptive web based systems with short
a collection of HTML tags organized in "one- description of our five layer top-down
size-fits-all"[3] maner. Different users however framework. In section 3, a complete ontology for
have various preferences even for the same web adaptive web based systems that comply with
site seen through longer period of time. In order our framework is presented. In section 4 some
to meet the above mentioned requirements, a rules and inference testing is being performed
system that rearranges its site's internal structure and section 5 concludes the paper.
and content needs to be created. These types of
web systems are said to be adaptive. 2. Related work
Adaptive Web-Based Systems can be
generalized into three main categories sumarized Most of the related work done concerning
as: Adaptive Web-Based Systems is on the design
- Adaptable Web Based Systems that arrange approaches. Some of the approaches can be
their content or structure based on abstracted as:
preliminary feedback from the user. This Complete data-centric approach where the
feedback is ensured in various ways like main focus is the data, its organization and
logging in or filling up web forms. These relationship. This approach is mostly used in
systems and approaches used are also called developing a model-driven context aware web
proactive systems or approaches [12]. applications as described in [5].
- Adaptive Web-Based Systems on the other The second approach is a web mining
hand tend to arrange their content or approach where data is exploited and
structure without direct "invasion" from the manipulated with no clear emphasis on
knowledge or in the best case, the knowledge third layer takes care of the information gathered
discovered needs additional explanation from a from users that accesses the page. The
domain expert [6,10]. information acquired here reside on the user
Finally the third approach is completely information repository, which stores user access
knowledge-centric, where data is abstracted and on the page's content in the form of web logs, as
generalized in higher levels of application logic. well as pattern repository that uses data mining
This approach is exemplified in [15,4,7]. techniques for extracting and accumulating user
The overhead approaches suffer from several access behavior. The adaptation layer consists of
drawbacks due to one way treatment of the set of rules on how to perform adaptation and
problem i.e. they are either completely data finally presentation layer rearranges the web
centric or completely knowledge centric. In [11] site's content or structure based on the rules from
we presented a top-down five layer framework adaptation layer.
that offers flexibility, expressiveness, Designing a complete ontology for adaptive
interoperability and modularity to address the web based systems is not a trivial task. Most of
above mentioned problems. It is coined as top- the ontology that exist today focuses mostly on
down because it starts with the data layer at the modeling the content and relationship of content
top and narrows down to the adapted content by on such systems. For example an ontology for
taking into consideration other layers that lay in the content of adaptive web based systems has
between. The layers are considered in such a way been presented in [2]. Its ontology consists of
to take into account both data and knowledge Domain Model and a Domain Dependent User
(concepts about data). This framework consists Model. However this model uses a deprecated
of five layers as depicted below in Fig. 1. user domain, considering that it does not take
into account the user access behavior, but only
the user view related directly to concept in the
domain model. As we will see in the next section
our proposed ontology completely incorporates
the user as a consisting part of adaptation.
Another ontology approach is presented in
Adaptive Personal Information Environment (a-
PIE) which is based on Fundamental Open
Hypermedia Model (FOHM) [9]. This approach
is focused more on enabling users to search
information based on ontologically defined
concepts, rather than creating an ontology that
will completely express an adaptive web based
system.
In the next section will give a complete
ontology that will fulfill the requirements of the
framework given earlier.

3. Modeling a complete ontology for

Adaptive Web Based Systems
Figure 1. A Top­Down Five Layer Framework 
For Adaptive Web Based Systems The main idea behind this ontology is to
envision a web site that will be consisted of
The first layer called a data layer is consisted loosely coupled atomic data units. Based on user
of loosely coupled atomic data units. The units preferences, these atomic units can be gathered
can be texts, multimedia files (audio, video or to create more composite unit (pages) that will be
animations) or any other describable content. presented in an adapted way. The ontology
These units represent the building blocks of a consists of concept domain and user domain. In
web page. The second layer represents a layer of concept domain we store atomic units as
concepts and concept relationships gathered from concepts as well as concept relationships. These
atomic units. The ontology that we present in this concepts have their own information weights
paper is designed to function in this layer. The (ratings) and can be aggregated into pages. The
generalized model of the ontology is illustrated reasoning in the core classes. This will be
as in Fig. 2 explained in section 4.
The user domain is consisted of concepts The AtomicUnit class can have a type, which
concerning the user like: user requests, sessions represents the general type of data that the
and user preferences. atomic unit can be consisted of, like: image,
multimedia (audio, video or animation) and text.
To completely describe the atomic unit ontology
additional classes are needed as well. For
example the Position subclass that describes the
position of atomic unit within a page and Link
subclass that creates the relationships between
these concepts.
Other classes in the core involves the Page
class which can be composed from several
atomic units, the Preference class consisted of
several sub classes namely general preferences
that the users might seek in a page, and Rating
class that gives the atomic unit an information
weight like: High, Medium or Low importance.
The complete class hierarchy as shown from
Protégé is presented in Fig. 3

Figure 2. The Generalized Model of the 
Complete Ontology

The steps involved in creating the ontology

for the above mentioned method encloses the
following:
1. Definition of concept classes and
hierarchies.
2. Definition of concept properties
3. Specifying the relationship between
concepts
4. Definition of concept instances for some
classes of the ontology
The ontology have been defined using OWL
DL1 Language, which allows enough
expressiveness as well as inference possibilities.
The definition of concept classes, relationships
and instances is done using Protégé ontology
editor and knowledge base system[14].

3.1. Definition of Concept Classes and

Hierarchies

The complete ontology consists of core class

and inference class. The core class consists of
five main classes namely: AtomicUnit, Page,
Preference, Rating and User. The inference
Figure 3. The Asserted model of the 
class called Visits will be used for testing the
Ontology
1
OWL DL – (DL – Description Logic)Represents a
sub language of Web Ontology Language(OWL) that
allows definition of logic rules and efficient reasoning
support.
3.2. Definition of Concept Properties and importance has been emphasized through
Relationships hasRating property which relates AtomicUnit
class to some Rating subclasses such as
The concept or class properties relates High, Medium or Low.
individuals from one class (domain) to another - Finally, three properties have been defined
class (range) creating this way a relationships concerning the user domain that creates
between concept classes. Without class relationships such as:
properties, the inference in the ontology would - User can have a request in a web page. The
not be possible, considering that these properties request can be a complete page or an atomic
create class restrictions between domains and unit and furthermore these requests can be
ranges. The complete list of concept properties, gathered into sessions. These relationships
with their domains and ranges used in our are described with hasRequest and
ontology is shown as in Table 1. hasSession property.
- And finally users can have a browsing
Table 1. Domains and Ranges of Class  preference which are being described by
Properties hasPreference property.

Property Name Domain Range 3.3. Specifying Concept Instances

hasAtomic Page AtomicUnit Instances represents individuals that belong to
hasType a class, these are like class members. We have
(subroperty of Page Type defined seven instances for the user class with
hasAtomic) the purpose of assigning them different
hasChild AtomicUnit AtomicUnit preferences to see the inference on the ontology.
hasPage AtomicUnit Page
4. Rules and Inference in the
hasParent AtomicUnit AtomicUnit Ontology
hasPosition AtomicUnit Page
hasPrefernce User Preference We have mentioned above that class
properties created relationships between
hasRating AtomicUnit Rating concepts. This relationship can be represented
hasRequest User Request by predicate logic[1]. If we would use the class
properties to link domain and range then we can
hasSession User Session
create several base predicates in the form:
property X i ,Y j 
The overall relationship of classes through X i={x1, x 2,. .. , xn } and
where both
class properties that has been used in this Y j={y 1, y 2,. .. , y n} , represent a domain and
ontology can be generalized as follows:
- Every atomic unit belongs to a certain page.
range of a class. From here we can create the
This relationship is presented through following base predicates for our ontology:
hasAtomic property. This property gives the
hasAtomic X Page ,Y AtomicUnit
possibility of synthesizing a page “on-the-
fly” based on atomic units. hasChildX AtomicUnit,Y AtomicUnit
- Each atomic unit can have a link from whom hasParent X AtomicUnit,Y AtomicUnit
it can be referenced and to whom it can hasPage X AtomicUnit,Y Page 
reference. This relationship is represented hasPosition X AtomicUnit,Y Page
through hasParent and hasChild property. hasPreference X User ,Y Preference
- Atomic units can have a certain position hasRating X AtomicUnit,Y Rating 
within a page. The position would involve hasRequest X User,Y Request
the right, left, center, header or footer region
hasSession X User,Y Session
of the page depending on the adaptation
scale. This relationships is described through
hasPosition property. Having these base predicates, we can further
- Atomic units can have importance within a infer by adding other rules. For example to
page concerning a certain concept. This express that an atomic unit can belong only to
one page, have only one rating (low, medium or Visits≡hasPreferencemin X User , Y Preference=3 
high) and can have exactly one position within a ,∧hasSessionmax X User ,Y Request=3
page we can write as follows:
Visits≡hasPosition X AtomicUnit, ∃Y Page 
atomicUnit hasPage X AtomicUnit ,Y Page=1, ,∧hasRating X AtomicUnit , ∀ Y Rating =HighRating
hasPosition X AtomicUnit ,Y Position =1 ,
hasRatingX AtomicUnit , Y Rating=1
      where:

Defining a page that might have some type of ∃ Y Page = { LeftPosition∨ Middle∨Header∨Footer}
atomic units and some certain number of it, can
be written as: The experienced visits on the other hand tend
to be more specific with well defined preferences
Page  hasType X Page ,∃ Y Type ,... that does not overpass two preferences per visit
hasAtomic X Page , max Y AtomicUnit=20 and the session tend are more long with an
average of 5 requests per session [8]. The rules
We have specified the maximum number of used for experienced visits are:
atomic units per page to 20 which is more than
enough. Visits≡hasPreferencemax X User , Y Preference=2∧
Other rules are as follows: hasSession
min  X User ,Y Request=3∧max X User ,Y Request=5
Preference hasRating X AtomicUnit ,Y Rating=1

Visits≡hasPosition X AtomicUnit ,∃ Y Page 

Rating  hasRating HighRating∨...
,∧hasRating
hasRatingMediumRating∨...
 X AtomicUnit , ∀ Y Rating= LowRating, MediumRating
hasRatingLowRating
where:
User  hasPreference X User ,∃Y Preference, ...
hasSession X User ,∃Y Session
∃ Y Page = { RightPosition∨Middle }

Request≡hasRequest X User , Y AtomicUnit=1∨..

The inference testing of these rules has been
hasRequestX User ,Y Page =1 
done by using Protégé's Fact++[13] inference
engine. The Fact++ uses the FaCT system (Fast
4.1. Inference in the ontology Classification of Terminologies) which
represents a Description Logic Classifier for
To test the inference in our ontology we have terminology classification. In OWL is being used
created new concept class called Visits with two for concept classification.
subclasses named NoviceVisits and From Fact++ reasoning two main
ExperiecedVisits respectively. The rules that we classification of concepts occur in our ontology.
have used for these two classes are that novice The first classification is that NoviceVisits and
visitors do not have a particular interest in the ExpericedVisits appear under AtomicUnits class.
topic but they browse randomly and the number Which is logical because in several rules for the
of requests is not very big. From average five visits above we have emphasized these visits
requests[8] per session they visit maximum two through position and ratings of atomic units
or three. Because they are unexperienced within a page as well as users preferences. We
visitors, atomic units should be putted in can conclude that the reasoner creates two types
different positions in a page as stated in the of atomic units in a page for two types of users.
Position subclass of AtomicUnit class, to address The second classification is the one that
their needs. This is helpful in avoiding the “lost happens under User class. We can see that under
in hyperspace” effect. this class we have ExperiencedVisits,
The rules used for Novice Visits are: NoviceVists and Preference listed which we did
not have in our original ontology. The reasoner
inferred to this conclusion based on the rules that
we have given above concerning the request that References
had some atomic units which had some rating.
The inferred classification classes can be seen [1] Antoniou G, Van-Harmelen F. A Semantic
below in Fig. 4. Web primer. MIT: MIT Press; 2nd Ed;2002.
[2] Bielikova M, Moravcik M. Modeling the
Content of Adaptive Web-Based System
Using an Ontology. First International
Workshop on Semantic Media Adaptation
and Personalization (SMAP'06); 2006
[3] Brusilovsky P, Maybury T-M. From
Adaptive Hypermedia to the Adaptive Web.
Journal of Communication of the ACM;
2002; 45: 31-33
[4] Carmagnola F, Cena F, Gena C, Torre I. A
Semantic Framework for Adaptive
Hypermedia Systems. In the proceeding of
"SWAP", the Second Italian Workshop on
Semantic Web Applications and
Perspectives; 2005
[5] Cerri S, Daniel F, Matera M, Facca F. Model-
Driven Development of Context-Aware Web
Applications. ACM Transactions on Internet
Technology (TOIT); 2007; 7
[6] Chibing G, Nordahl M. Building an Adaptive
Website Based on User Access Patterns.
Proceedings of the International Conference
on Cyberworlds (CW'05); 2005
[7] Cristea A. What can the Semantic Web do for
Figure 4. The Inferred model of the  Ontology Adaptive Educational Hypermedia?
International Peer-Reviewed On-line Journal
5. Conclusion and Future Work "Educational Technology and Society" 7(4),
Special Issue on Ontologies and the
In this paper we have presented a complete Semantic Web for E-learning, IEEE, LTSC;
knowledge representation for Adaptive Web- 2004; 40-58
Based Systems using ontologies. The ontology [8] Georgieva J, Raufi B, Ahmedi L, Ismaili F,
used comply with the requirements for Zenuni Xh. XML-Based Sesion
expressiveness, modularity and interoperability Reconstruction for Adaptive Web-Based
presented in Top-Down Five Layer framework. Systems. Computer Engineering; 2008; (3)
We have also created some rules and tested the [9] Maneewatthena, Th., Wills, G. B., Hall, W.
ontology inference in an inference engine and we Adaptive Personal Information Environment
saw that findings comply and can be used as based on the Semantic Web. Proceedings of
knowledge base for the second and third layer of the sixteenth ACM conference on Hypertext
our framework. and hypermedia HYPERTEXT'05; 2005;
The future work would involve creating a 214-216
data representation that will use the proposed [10] Perkowitz M, Etzioni O. Adaptive Web
ontology, finding the possibility of automated Sites: Examining the potential use of
creation of content-concept linking, using automated adaptation to improve web sites
adaptation algorithms and presenting a content in for visitors. Journal of Communication of the
presentation layer. It is worth mentioning that the ACM; 2000; 43: 152-158
above tasks are not trivial and require the [11] Raufi B, Georgieva J. Adaptive-Web Based
involvement of many disciplines like web Systems: Current Trends and Open Issues. In
mining, machine learning, semantic web and the proc. Of 4th International. Bulgarian-
many other related fields. Greek Conference on Computer Science
2008. Kavala, Greece; 2008
[12] Spiliopoulou M, Mobasher B, Berendt B, [14] The official Protégé Page; 2008.
Nakagawa M. A Framework for the http://protege.stanford.edu/ [17/01/2009]
Evaluation of Session Reconstruction [15] Xing W, Ghorbani A.A. Information
Heuristics in Web-Usage Analysis. Domain Modeling for Adaptive Web
INFORMS Journal on Computing; 2003; Systems. Proceedings of the
15(2): 171-190 IEEE/WIC/ACM International Conference
[13] The FaCT System. 2008. on Web Intelligence; 2005
http://www.cs.man.ac.uk/~horrocks/FaCT/
[18/01/2009]