WHAT ARE MAIN CONCEPTS IN AN OWL DOMAIN

ONTOLOGY?

Christian Kop

Institute of Applied Informatics, Alpen-Adria-Universitaet Klagenfurt

Universitaetsstrasse 65-67, 9020 Klagenfurt, Austria

Keywords: Focus of ontology, Ontology structure, Main concept.

Abstract: Whereas OWL is suitable for machine interpretation, it is hard to read for a human and it is hard to

understand what the real focus of the ontology is. This paper will discuss if measures based on the ontology

structure can help.

1 INTRODUCTION

Whereas OWL (McGuiness et.al., 2004) is suitable

for machine interpretation, it is hard for a human to

get an overview of the ontology focus in order to

decide if the content of the ontology is correctly

expressed by it’s name or it’s introductive

comments. For instance, does the name of an

ontology (e.g. pizza.owl) also reflect the ontology’s

content (it’s structure)? Such information is useful to

decide if the ontology or it’s elements are

appropriate in a similar domain.

In order to get this kind of information, the

ontology structure is searched for concepts (main

concepts) which seem to be important (Huang et.al.

2006).

Therefore the paper is structured as follows. In

the next section it will be discussed how main

concepts can be detected. Afterwards, possible

application scenarios are given in section 3. Section

4 finally gives a conclusion.

2 WHAT ARE MAIN CONCEPTS?

With regard to an ontology purpose there are always

concepts which have a richer description within an

ontology than others to underline their importance

for that purpose. Such descriptions might be the

involvement in a generalization hierarchy or

relationships to other concepts. These concepts are

called main concepts in this paper.

If these concepts can be detected, then they are

helpful to get an overview of the ontology structure.

Thus, the expectations of the human can be

compared with the ontology focus derived from the

structure.

2.1 Measures based on the Structure

Whereas, in (Huang et.al. 2006) only one measure

was used, this approach discusses four measures

namely: Weighted number of successors (wNS),

number of object properties (P), weighted number of

object properties (wP) and instances of an OWL

class (I). They reflect possible different ways how

main concepts can be modeled (Guarino, 1995).

Weighted Number of Successors (wNS). The

modeling construct often used in ontologies is

generalization. In (Bezerra et.al, 2009) the number

of direct children is counted. Alternatively, the

number of successors (NS) in the whole subclass

hierarchy of a certain concept is counted here. To

avoid that OWL classes at the top of the hierarchy

will always be the winners, the number of successors

are multiplied with a weighting (figure 1).

This weighting is defined as follows: The root

element (“Thing”) is weighted with 0. From the root

element the path to the concept is calculated as

“distance from root” (dfr). Additionally, the

maximum distance to a leaf from that concept is

calculated (maxdtl).

For a class which is a leave in the hierarchy,

maxdtl is defined as 1. The weighting factor is dfr /

maxdtl. The weighted number of successors for a

concept X then is: wNS(X) = NS * (dfr / maxdtl)

404

Kop C. (2009).

WHAT ARE MAIN CONCEPTS IN AN OWL DOMAIN ONTOLOGY?.

In Proceedings of the International Conference on Knowledge Engineering and Ontology Development, pages 404-407

DOI: 10.5220/0002294204040407

 SciTePress

c7c6

wNS(c4) = 4 * (1 / 2)

distance

from root

(dfr) = 1

maximum

distance to

a leaf of c4

(maxdtl) = 2

number of

successors of

c4 (NS) = 4

c7c6

wNS(c4) = 4 * (1 / 2)

distance

from root

(dfr) = 1

maximum

distance to

a leaf of c4

(maxdtl) = 2

number of

successors of

c4 (NS) = 4

Figure 1: Example for wNS calculation.

Number of Object Properties for a Domain Class

(P). An OWL object property points from the

domain class to the range class. In other words, the

domain is described with the range. Thus domain

classes are more likely main concepts. If a certain

class is involved as a domain in an object property,

then a counter is incremented by 1 for that class.

Weighted Number of Object Properties (wP). A

more refined version to count is to add the weighted

number of successors of the range class to the

domain class. However, if wNS for the range is 0

then wP is incremented by 1. In this case, it

degrades to P. With this strategy the importance

(weight) of the range is forwarded to the domain.

Instances of an OWL Class (I). Defining many

instances of a class can be an additional way to

describe a main concept.

2.2 Discussion

The measures in the previous section were tested on

6 ontologies which were found on the web site:

http://krono.act.uji.es/Links/ontologies/. Particularly

the well known pizza-, food- and wine-ontology as

well as an ontology of photography, of rheumatic

patients and a simple ontology of an university were

selected. These ontologies were chosen for the

following reasons: According to the filenames it was

expected, that they describe typical domain

ontologies. Furthermore these domain ontologies are

not too specific (e.g. very specific medical,

biological or technical domains) but domains, which

represent general knowledge also known to the

author. Hence it is possible to easily compare the

results of the measures with the author’s

understanding of the ontology.

Pizza Ontology. Using the wNS, Named Pizza

followed by Pizza, Pizza Topping, Vegetable

Topping, Cheese Topping, Domain Concept are the

best ranked concepts. Using wP, there are three

important concepts, namely Pizza Topping followed

by Pizza Base and Pizza. Pizza Topping and Pizza

Base are important concepts, since in the original

ontology, they are also domains of an object

property (Pizza Topping has Topping of Pizza and

Pizza Base is base of Pizza). The same holds for P

but now Pizza is top ranked because it is involved in

two object properties whereas the two others are

only involved in one object property as a domain

class. Using I, only Country appears on top of the

list. Hence this measure alone does not give good

hints for main concepts in this ontology. What can

now be said on the basis of three of the four

measures is that in fact the ontology describes Pizzas

and how they are made (e.g. made with some

Topping and some Pizza Base). The measures

reflect the author’s intuitive understanding and

expectations of this domain.

Food Ontology. The measures applied on the food

ontology return the following results. With the wNS

the three top ranked concepts are Meal Course,

Edible Thing and Consumable Thing. Then some

other concepts like Pasta, Seafood, Fish, Pasta with

red sauce

etc. follow. But these concepts did not get

that high score. Meal and Meal Course were also top

ranked in the wP measure. Consumable Thing got

the third position in the P and wP ranking. For I,

Oyster Shellfish, Sweet Fruit and some other

concepts were top ranked. Once again except of the

instance counter (I) the results look promising.

Wine Ontology. During determination of the

concept statistics one problem arose. The wine

ontology imports the food ontology. If also the

imported concepts from the food ontology are

measured, then the concepts of the food ontology are

top ranked and not the concepts of the wine

ontology. However if the wine ontology is

determined locally, then the following can be said:

According to wNS the concept Wine is the highest

ranked concept. The next concept is Loire.

Afterwards, Bordaux, Medoc, White Wine follows.

But these concepts do not have the same high score

like the concept Wine. Applying wP and P, Wine,

Region and Vintage were best ranked. Finally with

I, Winery, Region, WineGrape were top ranked,

followed by some other concepts. Here the first time

also the concepts ranked by their number of

instances make sense.

Photography Ontology. With wNS, Film,

Equipment, Lens, Light Sensitive Auto Focus,

WHAT ARE MAIN CONCEPTS IN AN OWL DOMAIN ONTOLOGY?

405

Camera, Filter, Physical Thing, Shutter Speed, Test

Lens, Exposure Parameter are on top of the list. No

values were determined for wP, P and I. The reason

is, that the object properties are totally independent

in this context. Also, no instances are defined for the

specified concepts. However, on the basis of wNS, it

can be said, that this ontology focuses on the

concepts named above.

Simple University Ontology. On the basis of wNS,

the top ranked concept in the university ontology

was the concept Module. Then the concept

Academic Rank follows. Afterwards the concepts

Age_group_simple_VT, Module Format, Salary

Range Value Type, Teaching Unit, Value Partition

appear in the list. Once again no statistics could be

derived for wP, P and I. The reasons are the same as

in the photography ontology. Here it was a little bit

surprising, that the concept Module was on the top

of the list. A detailed look on this concept showed

that Module is a Teaching Unit. Hence the top

ranked concepts fits with the idea of this ontology,

though it might be expected from the file name

“Simple University-01.owl” that the organizational

structure of the university is described. The

measures showed that this is not the case. In fact

concepts for “Teaching” are described. Also those

restrictions referring to external concepts defined in

another ontology refer to a resource “Teaching-1-

01.daml”. Thus the top ranked concept reflects the

main concepts of the ontology.

Patient Rheuma Ontology. The last examined

ontology was the patient/rheuma ontology. Although

it focuses on a special medical domain, it was

assumed that there are at least some top ranked

notions which are also known to non specialists (e.g.

“patient”, “rheuma”). Surprisingly, the top ranked

concepts are not patient or rheuma as it might be

expected from the name of the file

(“PatientRheuma.owl”). Instead, applying wNS the

concepts Ward, Number, Hospital, Medical

Organisation, Joint Inflammation and Physician are

firstly listed. Applying wP, Transport, Flight, List of

Hospitals, Diagnosis are the four top ranked

concepts. With the P measure, the ordering of the

first four concepts is slightly different: Transport,

Flight, Organization, and Diagnosis (resp. Patient

or Address which have the same result for P as

Diagnosis). The concepts Gene, Diagnosis, Address

and Patient have instances (I) but there are not so

many. Gene has only 2 instances. The others have

only 1 instance. Looking at these results the question

arose, why they do not match with the expectations.

Why is e.g. Number at a very good ranking position

in the wNS ranking? Why has Patient such a bad

ranking? What about “rheuma”? The reasons for that

can be found inside the ontology. Everything that is

a number (e.g. Booking Number, Credit Card

Number, Flight Number etc.) was subsumed to

Number. Patient does not have a deeper

substructure. Instead it is a subclass of Person and is

also a leaf in the taxonomy tree. The concept

Rheuma does not exist as such. Instead “Rheumatoid

Arthritis” is mentioned in the ontology. This once

again is a leaf in this local ontology. Whereas

Rheumatoid Arthritis references another external

ontology, Patient does not have such a reference.

Most surprising was the fact, that in the object

properties sections, many properties have the

domain class Flight or Transport. The concept

Patient is involved only in two object properties as a

domain class. Even together with its super class

(Person) the number of object properties was less

than the number of object properties defined for the

concept flight. The super class Disease can be found

in only one object property. A look on the number of

restrictions for a patient concept showed that Patient

was involved in two someValuesFrom restrictions.

Although it was a surprise, in fact the results gave a

good picture of the ontology structure, since the

concepts “patient” and “rheuma” are not really

described in detail in the local context of the file

“PatientRheuma.owl”. Instead the ontology

engineers focused more on the description concepts

like ward, hospital, flight, transport etc. Thus if

someone would like to (re)use detailed information

about patients, this ontology is not the best one for

doing it.

3 APPLICATION SCENARIOS

The results of the measures can be the basis for two

application scenarios.

These results can be used to generate a natural

language abstract (summary) of the ontology.

Strategies how to verbalize ontologies are described

in (Fuchs et.al. 2005), (Hewlett et. al. 2005) and

(Fliedl et. al 2007). In combination with the

described measures a summary can be generated if

only main concepts are verbalized.

Another application scenario is the mapping of

ontology elements to a conceptual database schema

or the support of information systems design

(Guarino, 1998), (Sugumaran 2006). Strategies for

mappings are described in (Vasilecas et.al. 2005)

and (Kalibatiene et.al. 2009). With the measures, a

KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development

406

selection of appropriate concepts can be made before

the mapping is applied on these concepts. Hence this

strategy would consider that the ontology and the

future database schema have a different scope and

focus, though belonging to the same domain.

4 CONLUSIONS

From the examination of the six ontologies, the

following can be learned: Except the instance

measure (I), the three other measures (wNS, wP, P)

give a good first impression about the focus of an

ontology structure. This can be observed also in the

case of the university and patient/rheuma ontology.

The weighted number of successor measure

(wNS) can be applied more often than the others,

since often taxonomies are used. Nevertheless also

the weighted property (wP) measure as well as the

property measure (P) are important. Especially for

users who want to re-use an ontology for conceptual

modeling, knowledge about the “relationships”

between concepts is interesting.

In order to give human readers the ability to

examine an ontology according to the measures, a

prototype was built (see figure 2 for the screenshot).

This prototype also allows browsing through the

ontology and it is a basis for the two application

scenarios.

Figure 2: Screenshot of the prototype.

In future also statistics of how often a certain

concept appears in a restriction (e.g.

someValuesFrom, allValuesFrom etc.) will be

analyzed.

REFERENCES

Bezerra, D., Costa, A., Okada, K., 2009, SwTO

(Software

Test Onotlogy Integrated) and its applicaton in Linux

Test. In Proceedings of the 3

International Workshop

on Ontology, Conceptualization for Information

Systems, Software Engineering and Service Science,

CEUR-WS, Vol 460, http://ftp.informatik.rwth-

aachen.de/Publications/CEUR-WS/, pp. 25 – 36.

Fliedl G., Kop C., Voehringer J., 2007. From OWL class

and property labels to human understandable natural

language. In Kedad Z., Lammari N., Métais E.,

Meziane F., Rezgui Y. (eds.), Proceedings of the 12th

International Conference on Applications of Natural

Language to Information Systems, NLDB 2007,

Lecture Notes in Computer Science (LNCS), Vol.

4592, Springer Verlag, 2005, pp. 156 – 167.

Fuchs, N.E., Höfler, S., Kaljurand, K., Rinaldi, F. and

Schneider, G., 2005. Attempto Controlled English: A

Knowledge Representation Language Readable by

Humans and Machines In Norbert Eisinger N. and

Maluszynski, J. (eds.): Reasoning Web, First

International Summer School, LNCS 3564,

Springer,2005, pp. 213-250.

Guarino, N. 1995. Formal Ontology, conceptual analysis

and Knowledge Representation. In International

Journal of Human-Computer Studies, Vol. 44, Issue.

5-6, 1995, pp. 625 – 640.

Guarino, N., 1998. Formal Ontology and Information

Systems. In Proceedings of FOIS’98, IOS Press, 1998,

pp. 3 – 15.

Hewlett, D., Kalyanpur, A., Kolovski, V., Halaschek-

Wiener, C., 2005. Effective Natural Language

Paraphrasing of Ontologies on the Semantic Web. In

End User Semantic Web Interaction Workshop,

CEUR-WS Proceedings, Vol. 172, 2005,

http://ftp.informatik.rwth-aachen.de/Publications/

CEUR-WS /

Huang, N., Diao, Sh., 2006. Structure-Based Ontology

Evaluation. In IEEE International Conference on e-

Business Engineering (ICEBE06), pp. 1- 6.

Kalibatiene, D., Vasilecas, O., Guizzardi, G., 2009.

Transformation Ontology Axioms to Information

Processing Rules – An MDA Based Approach. In

Proceedings of the 3

International Workshop on

Ontology, Conceptualization for Information Systems,

Software Engineering and Service Science, CEUR-

WS, Vol 460, http://ftp.informatik.rwth-

aachen.de/Publications/ CEUR-WS/, pp. 25 – 36.

McGuiness, D.L., van Harmelen F., 2004. OWL Web

Ontology Language Overview, http://www.w3.org/TR/

owl-features/

Sugumaran, V., Storey, V., 2006. The Role of Domain

Ontologies in Database Design: An Ontology

Management and Conceptual Modeling Environment.

In ACM Transaction on Data-base Systems, Vol. 31,

No. 3, Sept. 2006, pp. 1064 – 1094.

Vasilecas, O., Bugaite, D., 2005. Ontology-Based

Elicitation of Business Rules. In A.G. Nilsson, R.

Gustas, W.G. Wojtkowski, W. Wojtkowski, S.

Wrycza, J. Zupancic, Advances in Information

Systems Development: Bridging the Gap between

Academia & Industry, Vol. 2, Springer Verlag,

Heidelberg, 2005 pp. 795 – 805.

WHAT ARE MAIN CONCEPTS IN AN OWL DOMAIN ONTOLOGY?

407