DECISION SUPPORT SYSTEMS IN ONTOLOGY-BASED
CONSTRUCTION OF WEB DIRECTORIES
Marko Horvat, Gordan Gledec and Nikola Bogunović
Faculty of Electrical Engineering and Computing, University of Zagreb
Unska 3, HR-10000 Zagreb, Croatia
Keywords: Ontology, Ontology Alignment, Artificial Intelligence, Semantic Web, Web directory.
Abstract: The paper proposes an ontology-based mechanism for fully automated development of a Web directory’s
structure using the Semantic web as an underlying and integrating principle. Maintenance of a Web
directory is a time and resource wise consuming task. Moreover, there is always a realistic risk of the
structure becoming unbalanced, uneven and difficult to use to all except for a few users proficient in a
particular Web directory. By using ontologies to describe semantics of Web resources and Web directory’s
categories, and through the use of ontology mapping, it is possible to construct algorithms that can build or
rearrange the structure of a Web directory. Such applications are immediately helpful but also can be useful
in the more general problem of ontology sorting.
1 INTRODUCTION
In the last eight years we are witnessing rapid
development of the Semantic web and the related
spectrum of technologies. The initial paper by Tim
Berners-Lee (Berners‐Lee et al., 2001) introduced
the notion of universally described semantics of
information and services on the Web. The vision of
a Web as a shared common medium for data,
information and knowledge exchange, and
collaboration, fostered a wealth of development and
research. The Semantic web brought the power of
managed expressivity provided by ontologies to the
World Wide Web (WWW). Today the research in
Semantic web application is not largely focused on
the problem of ontologically-based Web directories.
So far only a handful or papers have been published
on the topic of combining ontologies and Web
directories (Choi, 2001; Kavalec and Svátek, 2002;
Mladenić, 1998). However, importance of Web
directories and their commonplaceness makes them
appealing for research.
The remainder of the paper is organized as
follows; the next chapter formally defines the
categories and the structure of Web directories. Web
directories construction scenarios are presented in
the third chapter, while the fourth chapter describes
an algorithm for their ontology-based construction.
Conclusion with outlook for future work is presented
at the end of the paper.
2 CATEGORIES AND THE
STRUCTURE OF WEB
DIRECTORIES
A Web directory is a structured and hierarchically
arranged collection of links to other web sites. Web
directories are divided into categories and
subcategories with a single top category, often called
the root category, or just the root. Each category can
have a provisional number of subcategories, with
each subcategory further subsuming any number of
other subcategories, and so on. Furthermore, every
category has a unique name and an accompanying
Uniform Resource Locator (URL), and can also
carry other associated information. Each category of
a Web directory contains a set of links to various
sites on the WWW, and a set of links to other
categories within the web directory. This basic trait
is the most important feature of a Web directory.
Associations between categories are arbitrary, but
there must be at least one path between any pair of
categories. Disjoint sets of categories are prohibited,
281
Horvat M., Gledec G. and Bogunovic N. (2009).
DECISION SUPPORT SYSTEMS IN ONTOLOGY-BASED CONSTRUCTION OF WEB DIRECTORIES.
In Proceedings of the 4th International Conference on Software and Data Technologies, pages 281-286
DOI: 10.5220/0002243102810286
Copyright
c
 SciTePress

as well as parallel links and self-loops. Each
category must have links to all its children, but can
also have links to other categories in the Web
directory which are semantically similar, or
otherwise analogous to the category (cross-links,
related links).
We will formally designate with C the set of all
categories in a Web directory; R will be the set of all
Web resources in a Web directory. One category
with unique identification number n is denoted c
n
.
Category has its own characteristic URL url and
member level l, where l is a natural number smaller
than or equal to the depth of a Web directory L
(Figure 1). The category c
n
must be a member of C.
C
n
is a subset of C that belongs to the category c
n
,
and R
n
the subset of R with Web resources that
belong to category c
n
. We formally describe
categories and structure of Web directories in
Uschold, 2003.
Figure 1: Schematic representation of a single category.
Mathematically speaking, Web directories are
simple rooted graphs (Sedgewick, 2001). Sometimes
the position of links within a category’s Web page is
prioritized, and in that case we are talking about
ordered and rooted simple graphs. The structure of a
Web category cannot be described as a tree because
more than one path can connect any of its two
categories: apart from paths which connect
parent/child categories, they can be associated with
ad hoc cross-links as in Figure 2.

Figure 2: Realistic Web directory with possible multiple
paths between two categories.
Although the categorization of a Web directory
should be defined by a standard and unchanging
policy this is frequently not the case. Web
directories often allow site owners to directly submit
their site for inclusion, even suggest an appropriate
category for the site, and have editors review
submissions. The editors must approve the
submission and decide in which category to put the
link in. However, rules that influence the editors’
decision are not completely objective and are thus
difficult to implement unambiguously. Sometimes a
site will fall in two or even more categories, or
require a new category. Defining a new category is
very sensitive task because it has to adequately
represent a number of sites, avoid interfering with
domains of other categories, and at the same time
the width and depth of the entire directory’s
structure has to be balanced. A Web directory with
elaborate structure at one end and sparse and
shallow at the other is confusing for users and
difficult to find quality information in. Furthermore,
after several sites have been added to a directory it
may become apparent that an entirely new
categorization could better represent the directory’s
content. In this case a part of directory’s structure or
even all of its levels have to be rearranged which is
again time and labor consuming task.
Therefore, recognizing the challenges implied by
the Web directory construction, and as well as their
overall importance, the paper’s authors are
motivated to design and develop a decision support
system – a computer-based intelligent agent – that
can support decision-making in this construction
process.
3 CONSTRUCTION SCENARIOS
The process of building Web directories has three
actors:
1. Web directory system (WDS)
2. Web directory administrator (WDA)
3. Administrator of a Web site listed in the Web
directory (WSA)
Ontology-based building process contains the
same three actors and represents a subset of the
general building process. This process includes three
main tasks, or actions, that have to be performed by
actors in order to construct a Web directory:
1. Semantics identification task (SIT)
2. Semantics assignment task (SAT)
3. Web directory addition task (WDAT)
ICSOFT 2009 - 4th International Conference on Software and Data Technologies
282

Semantics identification task is a process that
recommends which ontology class, or classes,
should be instantiated and assigned to a given Web
resource. Semantics assignment task is a process that
follows semantics identification, and actually
assigns a set of ontology classes to a resource.
Classes that are recommended and assigned don’t
necessarily have to be identical. If an actor has made
an error and recommended the wrong class, the actor
performing assignment can overrule his
recommendation. Finally, when a set of classes has
been assigned to a Web resource, it has to be added
to a directory. Web directory addition task decides
exactly where in a directory’s structure the new
resource will be placed. This is a complicated task
because it can involve creation of an entirely new
category, reshuffling and updating existing
categories (both horizontally and vertically within
the directory’s structure), or simply adding the
resource to an existing category. The order of these
tasks and their mutual interaction is described in the
following UML activity diagram (Figure 3).
Figure 3: Main tasks in ontology-based building of Web
directories.
Construction process scenarios can be divided in
two groups:
1. Dominantly autonomous scenario (AUTO)
2. Man-In-The-Loop dominant scenario (MIL)
Each scenario has several possible variations or
sub-scenarios. Scenarios are distinguished by the
level of human participation. Sub-scenarios describe
the roles of the actors involved.
Utilization of human intelligence in majority of tasks
is presumed in MIL scenario, while in AUTO
scenario the Web directory computer system
performs more tasks than human actors. In an ideal
AUTO scenario the computer executes all tasks
independently. Table 1 depicts all scenarios and
their variations with respective grades of positivity.
Table 1: Allocation of actors and task in ontology-based
construction of a Web directory.
1
Tasks Roles
MIL scenario WSA WDA WDS
SIT ++ + n/a
SAT + +++ ++
WDAT - +++ +++
AUTO scenario WSA WDA WDS
SIT + + ++
SAT - +++ ++
WDAT - + +++
By following the highest grades in each scenario it is
possible to determine the best actor for each task.
Sequences of the best choices for each task are
shown in UML diagrams in Figure 4 and Figure 5.
Data in the table, temporally structured in the
diagrams, reflects the “Best Practice” experience
gathered during 15 years of administrating the
Croatian Web directory (http://www.hr/) (Gledec et
al., 1999).

Figure 4: UML sequence diagram with the selection of
best actors in the MIL scenario.
As can be seen in MIL scenario, WSA is the best
actor to perform SIT, and WDA for SAT. In this
scenario SIT is intentionally performed only by a
human actor. WDAT can be executed equally good
by WDA or WDS, but it would be wrong to leave
this task to WSA. The reasoning behind allocation of
actors in this scenario is that WSA is the least
dependable actor and its contribution is the most
likely to be subjective and erroneous. The task will
be most successfully performed by WDA, but it
would be inefficient and wrong to give all tasks only
1
Sub-scenario grades: +++: the most acceptable, ++
favorable, + positive, –negative/unfavorable scenario,
n/a not applicable.
DECISION SUPPORT SYSTEMS IN ONTOLOGY-BASED CONSTRUCTION OF WEB DIRECTORIES
283

to WDA. After all, one of the principal goals of the
proposed system is to alleviate the burden of Web
directory administration from the amenable
personnel, and not to leave them with an equally
difficult job. The best option is to allocate SIT to
WSA and to leave the final decision about semantics
to WDA who is the most knowledgeable and
dependable actor of the three.

Figure 5: UML sequence diagram with the selection of
best actors in the AUTO scenario.
Much the same reasoning is reflected in the
AUTO scenario; however the importance of WDS in
this scenario is emphasized. Thus, WDS is the
optimal choice for executing SIT and WDAT.
Again, WDA will perform the final assignment of
ontologies to resources (i.e. SAT) to reduce possible
errors to a minimum. In this scenario it was
determined that it would be negative to let WSA to
execute SAT and WDAT since WDA or WDS can
perform a better job at this tasks. In this scenario
WSA and WDA are equally suitable to execute SIT.
If SAT is also given to WDS then the Web directory
building system is fully automated and autonomous.
4 ONTOLOGICALLY BASED
CONSTRUCTION
If it is possible to assign ontology to a Web resource
and execute semantics identification and semantics
assignment tasks as outlined in the previous chapter,
it is also possible to define an ontology-based
algorithm for automated construction of a Web
directory structure. Such algorithm performs all
tasks outlined in Figure 2. The algorithm’s input are
links to Web resources that are being added to the
Web directory, and output is schema of the
directory. Schema can be represented in a number of
ways, e.g. as a markup language, or additionally the
algorithm can use the schema to automatically build
the directory by writing and storing necessary static
and dynamic Web files like HTML, JavaScript,
PHP, etc.
In order to be able to define the described
algorithm we will assume that we have at a
disposition function sem that takes a resource
i
r
∈
R
and from its semantic content builds an
ontology
i
o
∈
O where R and O are sets of all
resources and ontologies, respectively.
:sem →RO (1)
The function sem builds an ontology from a
resource, i.e. it performs semantics identification and
semantics assignment tasks by creating a solid
representation of an abstract property. This property
can be described as informal and explicit on the
semantic continuum scale (Uschold, 2003) and its
technical realization is strictly formal. Operations of
the function sem can be performed by a computer
system or a domain expert, in which case we talk
about automatic or manual ontology construction,
respectively. The function sem is described in detail
(Horvat et al., 2009).
The basis for the algorithm construction process
is definition of category c
i
and its set of ontologies
O
i
as a unified pair (c
i
, O
i
). In acquiring O
i
the
algorithm uses the function sem and treats c
i
as a
Web resource. The input is a set of Web resources R
and the algorithm picks one resource r
i
at the time,
translates in into an ontology o
NEW
and calculates
distance between o
NEW
and every ontology in the
Web directory O looking for the closest. Categories
are compared using their member ontologies. At
each moment wd has n categories and a new
category has index n+1.
Pseudocode for ontology-based construction of Web
directories
add (c
1
, Ø)
FOR each r
i
in R (i = 1,…,m)
create ontology sem : r
i
→ O
NEW
IF K(C) = 1 THEN
create category (c
NEW
, O
NEW
)
add r
i
in c
NEW
add c
NEW
in wd as
l+1
n+1
c
ELSE
find the closest category (
l
n
c
,
n
O ) to
O
NEW
ICSOFT 2009 - 4th International Conference on Software and Data Technologies
284

d = dist(O
NEW
, O
n
)
CASE OF d
> mindist
V
:
create category (c
NEW
, O
NEW
)
add r
i
in c
NEW
add c
NEW
in wd as
l+1
n+1
c
> mindist
H
:
create category (c
NEW
, O
NEW
)
add r
i
in c
NEW
add c
NEW
in wd as
l
n+1
c
OTHERS:
add r
i
in
l
n
c
END CASE
END CASE
END LOOP
The most significant aspect of the algorithm is
reliance on ontologies and ontology aligning
methods in order to measure similarity between
ontologies and determine their mutual distance. The
similarity measure sim : C
2
→ [0,1] between the two
categories
12
,cc∈C
and a distance function
() ()
12 12
, 1 / ,dist c c sim c c=
is defined elsewhere
as in (Staab and Studer, 2004; Ehrig and Sure,
2004). The algorithm uses two constants in a
predefined metric; minimal horizontal semantic
distance (mindist
H
) and minimal vertical semantic
distance (mindist
V
) as thresholds in the category
addition process. When a new category c
j
is being
added and category c
i
already exists in wd if
dist(c
i
,c
j
) > mindist
H
then the algorithm will add c
j
as
a new category of wd. Likewise, if
dist(c
i
,c
j
) > mindist
V
then c
j
will be added in a new
level of the directory wd, below c
i
. If
dist(c
i
,c
j
) <= mindist
H
AND dist(c
i
,c
j
) <= mindist
V
the algorithm will merge semantics of c
j
and c
i
incrementing initial ontology of c
i
. Therefore, the
thresholds are used in deciding whether it is
necessary to add a new category in the directory’s
structure or to use an existing category. Also, the
thresholds indicate where to add a new category: in
the same level next to an existing category or below
it.
The algorithm has two main branches. The first
branch recognizes one special case when cardinal
number
Κ of all categories C in wd is 1, and the
second branch processes three cases with cardinality
of categories greater than 1. If
() 1CΚ= then l = 1
and only the root category has been added to wd. In
this case it is not necessary to calculate the distance
between ontologies and a new category can be
immediately constructed. If
() 1CΚ> there are more
categories, not just the root, and links to Web
resources are assigned to the semantically closest
categories. New categories are created if needed.
The single root node does not have a set of links
to Web resources (R
1
= Ø) and it is assigned to an
empty ontology (c
1
, Ø), however the algorithm can
be modified so it allows predefinition of main topics
in a Web portal or Web directory according to the
desired administrating policies.
The proposed algorithm is simple because it
represents the direct and the most obvious
implementation of an ontological principle in Web
directory construction. Categories cannot be
mutually prioritized, and the end structure is
completely dependent on the order of links to Web
resources which are the algorithm’s input.
Furthermore, there is no back-tracking or iterative
optimization. For these reasons the algorithm may
also be called basic or elementary, since all other
ontology-based algorithms should provide better
results. It could be used as an etalon for comparison
of different algorithms for construction of Web
directories.
Execution of this algorithm can be assigned to
different roles in MIL and AUTO scenarios (see
Table 1). For example, a part of the algorithm, like
SAT can be given to human experts (WDS or WDA)
and other tasks – SIT and especially WDAT – can
be executed by an intelligent agent (WDS). Different
assignment will yield diverse results and this
presents an interesting topic for further study and
experiments.
5 CONCLUSIONS
The primary goal of this paper was to envision and
demonstrate a method for automatic construction
and maintenance of Web directories content and
structure.
We would advise caution in joining Web
directories with ontologies and the Semantic web
paradigms. Structures of Web directories are often
biased and influenced by the contributors of
resources. Administration of a large directory is an
overwhelming task prone to errors. Therefore, it may
be better to construct ontologies from smaller
directories or from directories with rigid
administrative policies. The former type of
directories is more numerous than the latter, but they
will also offer less information and in a more
specialized area.
DECISION SUPPORT SYSTEMS IN ONTOLOGY-BASED CONSTRUCTION OF WEB DIRECTORIES
285

In the future work we would like to expand the
initial system and build a hard general ontology
which would efficiently encompass smaller
ontologies of individual categories and provide a
unitary base for ontology matching throughout the
Web directory. Furthermore, we would like to test
the upgraded system in real-life situations and use it
regularly as a decision support expert system in
maintenance of a large Web directory. In the near
future we are planning to validate the system and
evaluate its features by implementing it within the
Croatian Web directory.
REFERENCES
Berners Lee, T., Hendler, J., Lassila, O.: The Semantic
Web. Scientific American, 284, 5, 34—43 (2001)
Choi, B.: Making Sense of Search Results by Automatic
Web-page Classifications. Proc. of WebNet 2001 --
World Conference on the WWW and Internet, 184—
186 (2001)
Kavalec, M., Svátek, V.: Information Extraction and
Ontology Learning Guided by Web Directory. In:
ECAI Workshop on NLP and ML for ontology
engineering, Lyon, (2002)
Mladenić, D.: Turning Yahoo into an Automatic Web
Page Classifier. In: Proc. 13th European Conference
on Artificial Intelligence, ECAI’98, 473—474 (1998)
Uschold, M.: Where are the Semantics in the Semantic
Web? AI Magazine, 24, 3, 25—36 (2003)
Horvat, M., Gledec, G., Bogunović, N.: Assessing
Semantic Quality of Web Directories Structure.
Submitted for publication in ICCCI 2009, Wroclaw,
Poland (2009)
Sedgewick, R.: Algorithms in C++, Third Edition.
Addison-Wesley (2001)
Gledec, G.; Jurić, J.; Matijašević, M, Mikuc, M.:
WWW.HR - Experiences with Web-server
development and maintenance. Proc.of the 18th
International Conference on Computers in
Telecommunications, 22nd International Convention
MIPRO'99, 93—96 (1999)
Staab, S., Studer, R.: Handbook on Ontologies.
Handbooks in Information Systems, Springer (2004)
Ehrig, M., Sure, Y.: Ontology Mapping – An Integrated
Approach. In: Proceedings of the First European
Semantic Web Symposium, 3053, 76—91 (2004)
ICSOFT 2009 - 4th International Conference on Software and Data Technologies
286