A REFERENCE ONTOLOGY BASED APPROACH FOR SERVICE

ORIENTED ONTOLOGY MANAGEMENT

Shuying Wang, Jinghui Lu and Miriam A. M. Capretz

Department of Electrical and Computer Engineering, Faculty of Engineering

University of Western Ontario, London, Ontario, N6A 5B9, Canada

Keywords: Service Oriented Architecture, Ontology Management.

Abstract: To establish effective information exchange among applications in a distributed B2B environment, the

business participants are not only required to share their functions or service interfaces, but in many cases,

they also need to exchange their data models. Ontology, as a popular semantic form of knowledge

representation, can be used to represent data models, thus allowing applications to locate and integrate these

models in a more intelligent way. In this paper, we introduce a reference ontology based approach for

service oriented ontology management. Specifically, STAR, a domain specific reference ontology, is built

and used for the experiments in a real life case. Furthermore, in order to validate and evaluate our approach

and implementation, a prototype system is developed to provide ontology deploying, browsing and mapping

operations on a service-oriented mechanism. Our experiments have provided promising results, which are

consistent with our original ideas of managing ontologies and optimizing ontology mappings to facilitate

data interoperability in a distributed environment.

1 INTRODUCTION

In Business-to-Business (B2B) applications, the

interoperability of heterogeneous data sources is an

important issue that is widely recognized in

information technology intensive organizations. To

establish effective information exchange among

applications, the business participants are not only

required to share their functions or service

interfaces, but in many cases, they also need to

exchange their data models. The traditional

message-based approaches (Hohpe and Woolf,

2003) require developers to retrieve data models

through messages and then to perform a one-to-one

mapping in order to identify and characterize

relationships between the models of two

applications. However, it is a major challenge to

create and maintain thousands of mappings for these

models. Furthermore, in order to share their models,

each application needs to publish its data model in a

location where other applications can easily locate

and retrieve the related models for information

exchange.

As the core of the semantic web, ontology is the

representation of knowledge in a certain domain.

Representing data models by ontologies and

mapping ontologies among the semantic resources is

an important approach for achieving semantic data

interoperability. At the same time, service oriented

architecture (SOA) is a key technology for

supporting interoperability among information and

processing data model interoperability.

Consequently, the significant potential of combining

SOA and ontology provides a promising solution to

improve semantic interoperability. For example, a

well-defined mapping process can be considered as a

component that provides a mapping service, which

can be implemented with various applications.

In this paper, we propose a reference ontology

based approach to support the interoperability of

heterogeneous data sources. The main idea of our

methodology is to make use of background

knowledge in an industry domain to enhance the

performance of alignment. Specifically, terms from

the different data sources are first mapped to

intermediate terms defined in the reference ontology,

and then their mapping is deduced based on the

semantic relation of the intermediate terms.

Furthermore, in order to examine our approach, five

experiments are designed to cover a comprehensive

validation of the ontology mapping strategies. We

examine these experiment results from generic

Wang S., Lu J. and A. M. Capretz M.

A REFERENCE ONTOLOGY BASED APPROACH FOR SERVICE ORIENTED ONTOLOGY MANAGEMENT.

DOI: 10.5220/0002787600690074

In Proceedings of the 6th International Conference on Web Information Systems and Technology (WEBIST 2010), page

ISBN: 978-989-674-025-2

ontology mapping to specified case study and

provide analysis of the proposed ontology mapping.

Section 2 introduces the related work on current

research. In Section 3, it introduces reference

ontology based mapping approach and the system

architecture. In Section 4, we design five

experiments aiming to evaluate the proposed

approach. Lastly, Section 5 presents the conclusions

and outlines a number of directions for future work.

2 RELATED WORK

Our work aims to extend the principles of the

ontology mapping approach as well as the emerging

Web Services standards in order to support the

manageability and interoperability of heterogeneous

data sources. A fundamental problem with ontology

mapping involves the integration of heterogeneous

data sources, which has been extensively researched

in the last two decades (Rahm and Bernstein, 2001).

Some research approaches (Sabou et al, 2006,

Aleksovski et al, 2006) have considered the use of

external background knowledge as a way of

obtaining semantic mappings between syntactically

dissimilar ontologies. WordNet is one of the most

frequently used sources of background knowledge.

The literature (Li et al, 1995) shows that WordNet

has been used successfully for word sense

disambiguation algorithms in other contexts,

particularly in text. Moreover, SUMO (2009) has

initially been created and further developed to

facilitate data interoperability, information search

and retrieval, and automated inference.

A substantial amount of literature has been

published about the combination of SOA and

ontology improving semantic interoperability (Staab

and Studer, 2004). MAFRA (Silva and Rocha, 2003)

is a toolkit used to maintain an ontology mapping

system and provides support for ontology mapping

tasks, such as the automatic specification of

semantic relations, negotiation and evolution.

Additionally, a distinct project proposed by Korotkiy

and Top is known as Onto-SOA (Korotkiy and Top,

2006). Onto-SOA integrates ontologies and SOA to

provide a mechanism for representing and exploiting

both the conceptual and behavioral domain aspects.

Specifically, it employs an ontology-based domain

model as a direct input to a service and enables the

exchange of messages between a service and its

consumer.

3 A REFERENCE ONTOLOGY

BASED APPROACH FOR

SERVICE ORIENTED

ONTOLOGY MANAGEMENT

3.1 Reference Ontology

In addressing the interoperability of heterogeneous

data sources, our approach relies on the reference

domain ontology as a semantic bridge between

different data models. The basic process in this

approach first aligns the concepts of the

corresponding data models involved in the business

process with the reference domain knowledge. Next,

we use the semantic information from this reference

knowledge to infer relationships between the

models. Lastly, the relationships are utilized to

induce an indirect set of mapping pairs and to

generate the required correspondences between data

models.

Figure 1: A Reference Ontology-Based Approach for

Ontology Mapping.

In our approach, we develop local ontologies to

represent different data models and reference

ontology as the semantic bridge between local

ontologies. The reference ontology represents the

shared vocabulary of a domain and defines basic

terms that can be combined to describe more

complex semantics in the local ontologies. For

example, in Figure 1, we show a local ontology Oa,

for the data model of Application A and another

local Ontology Ob, for the corresponding data model

of Application B. Accordingly, these local

ontologies are used for exchanging data between

Application A and Application B. The reference

ontology Or represents the set of basic domain terms

that provide a semantic link between different data

models. Consequently, for concept a in local

WEBIST 2010 - 6th International Conference on Web Information Systems and Technologies

ontology Oa and concept b in local ontology Ob,

there are corresponding concepts a' and b' in the

reference ontology.

The concrete steps of this approach are outlined

below:

1. The reference ontology containing concepts,

Or(a') and Or(b'), correspond to Oa(a) and Ob(b) in

local ontologies;

2. For each pair of concepts:

y If Or(a') is

equivalentClass to Or(b') in the

reference ontology, then Oa(a) from ontology

A can be inferred as equivalent to Ob(b) from

ontology B;

y If Or(a') is

subClassOf Or(b') in the

reference ontology, then Oa(a) from ontology

A can be inferred as child of Ob(b) from

ontology B;

y If Or(a') is

sameAs Or(b') in the reference

ontology, then Oa(a) from ontology A can be

treated as synonym of Ob(b) from ontology B;

y If Or(a') is

differentFrom Or(b') in the

reference ontology, then Oa(a) from ontology

A can be inferred to be different from Ob(b)

from ontology B.

Moreover, if no relationship is found, then no

mapping is inferred.

3.2 STAR Ontology

In our case study, we are using the STAR ontology

as the mediator, which contains rich reference

knowledge for performing the intermediate

mapping.

Table 1: The Metrics of the STAR Ontology.

Metrics Count

Class 122

Object Property 117

Data Property 1168

Subclass Axioms 185

Object Property Domain Axioms 117

Object Property Range Axioms 115

Data Property Domain Axioms 1081

Data Property Range Axioms 1118

Standards for Technology in Automotive Retail

(STAR, 2009) is a non-profit, unionized

organization whose members include dealers,

manufacturers, retail system providers and

automotive-related industrial organizations. The goal

of the STAR organization is to use non-proprietary

information technology (IT) standards as a catalyst

in fulfilling the business information needs of

dealers and manufacturers. Using the STAR

metadata, we have developed the STAR reference

ontology in order to gain a high level of detailed

knowledge from the automotive retail industry

domain and therefore to facilitate the interoperability

of automotive retail applications.

In general, the STAR ontology is based on the

terminology in the automotive retail industry.

Specifically, it is formalized in OWL-DL. Currently,

it describes 1592 lexical terms; the ontology metrics

are listed in Table 1. STAR ontology covers

concepts in the automotive retail domain and,

accordingly, it is structured in six different

categories. These categories include General,

Dealer, Customer, Parts Management, Vehicle

Management and Sales and Vehicle Repair and

Service. Together, these six categories represent the

main organizational structure of STAR.

3.3 System Architecture

As depicted in Figure 2, our system is divided into

three layers. These layers include the semantic layer,

the service layer and the access layer. At the

semantic layer, the semantic and expressive

descriptions are used to describe the data models.

Data model ontology sources, also known as local

ontologies, are developed to represent different data

models. Subsequently, the information

corresponding to the ontologies, such as the business

entity name and the business process name, are

published using Web Services.

At the service layer, the regular Web Service

technologies are utilized. In particular, we are using

SOAP for messaging, WSDL for service description,

and the service registry for publishing, discovering,

and retrieving data models. Finally, a mapping

engine is used to execute ontology mapping.

For implementing the service registry, we have

also developed three key Web Services:

• The Publishing Service allows web users to

submit their ontologies and other related

information to the web. The submitted

information includes the provider of the

ontologies and the business process to which the

ontologies are applied.

• The Discovery Service executes a search based

on the information given by business processes

or ontology providers. When a user submits a

request for acquiring a certain ontology, the

service returns the available ontology list as the

search results.

A REFERENCE ONTOLOGY BASED APPROACH FOR SERVICE ORIENTED ONTOLOGY MANAGEMENT

• The Mapping Service provides the functionality

of ontology mappings. The mapping engines are

externally developed and imported into system.

Figure 2: System Architecture.

The access layer contains the Web Portal, which

enables users to access the available Web Services.

Users can also execute specific functions, such as

accessing the publishing ontology by using the

publishing services, searching and retrieving

ontologies via the discovery service, and performing

ontology mapping through the mapping service.

4 EXPERIMENTS

4.1 Experiment Design

A web-based prototype called Service Oriented

Ontology Management Framework (SOOMF) is

implemented in order for the end user to manage

ontologies. The implementation uses Struts and

Spring architecture. Apache Tomcat is used as the

container for the development and deployment in

Web services. Also, Jena is used as the OWL parser.

Based on the prototype, we validate the accuracy

improvement for the proposed mapping approach.

We classify five experiments into two categories; the

first category, which includes the first four tests, is

known as generic experiments. These experiments

are used to validate our reference ontology-based

approach in comparison to other existing mapping

methodologies (experiments 1 to 4). The second

category includes the specified experiment,

Experiment 5, where the newly developed STAR

ontology will be examined for the improved utility

and to a broader adoption of ontology mapping in

the automotive retail domain.

In our study, we design two sets of experimental

ontologies. First, in our generic experiments, we use

the data set that has been used in the Information

Interpretation and Integration Conference (I3CON,

2003) experimental ontologies. The second set of

ontologies is created to present specific data models

involved in certain business processes for

specialized terms in the automotive retail industry.

This data set is used in Experiment 5, which is

treated as the specified experiment. In this set, the

local ontologies are created to represent the data

models in the automotive retail industry, while the

SUMO (2009) and STAR ontologies are used as the

reference ontologies.

Experiment 1 - Terminological Approach. In

this experiment, we use the typical Terminological

approach, which uses a combination of lexical and

structural correspondence between source and target

ontologies to compare strings. The correspondence

of terms mapped from the two ontologies is

generated directly by the terminological mapping,

which results in a list of equivalent terms of pairs.

We expect that the result of this experiment assist in

constructing the evaluation baseline.

Experiment 2 - WordNet Hierarchical

Distance. WordNet is a freely available English

lexical database whose design is inspired by current

psycholinguistic theories of human lexical memory

(Li et al., 1995). This experiment is to implement the

mapping by using WordNet as the thesaurus for

calculating the hierarchical semantic distance.

Experiment 3 - The Combination of the

Terminological Approach and WordNet

Thesaurus. Experiment 3 combines the

Terminological approach and the use of WordNet to

refine the mapping results. We believe that

combining the Terminological approach and the

WordNet Thesaurus can overcome some of the

lexical limitations and improve the mapping

performance.

Experiment 4 - Reference Ontology-based

Mapping using SUMO as the Generic Reference

Knowledge. This experiment is to assess the value

of using standard upper ontologies, thus we utilize

the reference knowledge as an ontological bridge

that indirectly infers mapping between local

ontologies. SUMO is used as reference ontology.

Experiment 5 - Domain Specific Reference

Ontology based Mapping. In this experiment, we

utilize the reference ontology as a source of

background information. Terms from the two local

ontologies are first mapped to terms in STAR, and

subsequently, their mapping is deduced on the basis

of the semantic relation between the terms. The

WEBIST 2010 - 6th International Conference on Web Information Systems and Technologies

experiment includes three steps. In Step 1, we utilize

the Terminological approach to generate a mapping

baseline. Step 2 introduces SUMO as the reference

ontology, showing the mapping result of this generic

reference ontology. In Step 3, STAR is used to

facilitate the attempt to capture correspondence in

the automotive retail industry.

4.2 Measurement

Basically, a threshold score is a lower limit for the

similarity score of two concepts that belong to the

respective source and target ontologies and that will

be treated as mapping pairs. For instance, if pairs of

mapping results produced have a threshold score of

0.60, indicating that the two concepts are considered

as mapping pairs if the similarity score between

them is greater than or equal to 0.60.

The effectiveness of the mapping approaches can

be measured by the precision of the mapping results.

We define the mapping accuracy as the ratio of

correct mappings N to the number of discovered

mappings M. The formula in Equation 1 shows the

mapping precision as a percentage value.

Equation 1: Mapping Precision.

4.3 Generic Experiment Analysis

The generic experiments consist of Experiments 1,

2, 3, and 4. Accordingly, Figure 3(a) demonstrates

the mapping precision arranged by the threshold

value of similarity, where the left side of the graph

displays the lowest value; the right side shows the

highest value; and the vertical line illustrates the

precision. The Terminological approach in

Experiment 1 is presented as the baseline experiment

in order to evaluate the other mapping results.

Specifically, we are attempting to directly discover

mapping pairs between the local ontologies.

The results of the four generic experiments are

presented in Figure 3(b). In comparison to the

Terminological approach used in Experiment 1, the

WordNet Distance approach in the second

experiment generates more mapping pairs.

For example, when the threshold is 0.8, the

WordNet Distance method (Experiment 2) returns

68 mapping pairs and among them, 16 are correct as

the precision is only 24%, whereas in the

Terminological approach (Experiment 1), the result

is 29 mapping pairs with 16 correct mappings, and

the mapping precision increases as 55%. Overall, the

results indicate that the mapping precision of the

WordNet Distance approach is not necessarily more

effective than the Terminological approach.

Furthermore, the federated mapping approach of

Terminological and WordNet in Experiment 3

demonstrates that the mapping results are more

effective than they are in the two approaches used in

Experiments 1 and 2 respectively, and moreover, a

lot of incorrect mapping results are eliminated in

Experiment 3. For example, when the threshold level

is 0.6, the mapping precision of the federated

approach is 67%; in comparison to the 16% of the

Terminological approach.

We also obtain a high mapping precision when

using the reference ontology based approach in

Experiment 4. However, we also observe while the

similarity threshold is increased, the mapping pairs

are reduced. For example, the number of mapping

pairs is 14 at threshold 0.2, while this number is

reduced to 1 at a threshold of 0.7. This drastic

decrease occurs because the terms used in

experimental ontologies do not have corresponding

definitions in SUMO, and therefore, most terms in

local ontologies cannot be bridged by the reference

ontology. Therefore, in order to overcome this

disadvantage, the appropriate reference ontology

needs to be selected prior to mapping in a specific

domain. Moreover, in order to obtain a more

effective mapping result, the pre-selected reference

ontology should include as many terms as possible.

4.4 Specified Experiment Analysis

In comparison to the traditional Terminological

approach, the domain specific reference ontology

results in improved mapping precision at the same

threshold level. For example, Figure 4(a) shows that

at threshold level 0.2, the mapping precision of

using STAR is 90%. Compared to the 12% achieved

in the Terminological approach, the precision with

STAR is increased by 78%.

The experiment also proves that domain-specific

information can help to improve the mapping results

for the reference ontology based approach. For

example, when using either SUMO or STAR, the

mapping precision curves are very similar; however,

when the threshold level is increased from 0.2 to 0.8,

STAR is more effective than SUMO. In particular,

as shown in Figure 4(b), using SUMO reduces the

mapping pairs from 41 to 3, whereas using STAR

only decreases the pairs from 59 to 34. This

discrepancy occurs since the increased inclusion of

terms in the STAR, has a significant effect on the

mappings. In particular, the resulting similarity

A REFERENCE ONTOLOGY BASED APPROACH FOR SERVICE ORIENTED ONTOLOGY MANAGEMENT

scores of correct concept pairs are increased, therefore demonstrating that the domain specific

Figure 3: The Mapping Results of Generic Experiments. Figure 4: The Mapping Result of Specified Experiments.

specific reference ontology can increase the amount

mapped terms and thus lead to more meaningful

mappings.

5 CONCLUSIONS

The goal of our research is to show the feasibility

and potential advantages of using a service-oriented

mechanism to build an ontology management

framework and using reference ontology as

background knowledge for ontology mapping. The

implementation and experiments based on real world

case have provided positive results, which are

consistent with our original ideas of ontology

management and mapping in the distributed

environment. Our future work includes building a

new mechanism to facilitate reference ontology to

discover more mapping pairs.

REFERENCES

Aleksovski, Z., Klein, M., ten Kate, W., Harmelen, F.

van., (2006). “Matching Unstructured Vocabularies

using a Background Ontology”, in Managing

Knowledge in a World of Networks, Springer-Verlag.

Hohpe, G., Woolf, B., (2003). Enterprise Integration

Patterns: Designing, Building, and Deploying

Messaging Solutions, Addison Wesley.

I3CON, (2003). http://www.atl.external.lmco.com/projects

/ontology/i3con.html

Korotkiy, M., Top, J., (2006). “Onto-SOA: From

Ontology-enabled SOA to Service-enabled

Ontologies”, Proceedings of the Advanced Int'l

Conference on Telecommunications and Int'l

Conference on Internet and Web Applications and

Services, IEEE Computer Society, Washington, DC,

USA, pp.124-124.

Li, X., Szpakowicz, S., Matwin, S., (1995). “A WordNet-

based Algorithm for Word Sense Disambiguation”,

Proceedings of IJCAI-95. Montréal, Canada.

Rahm, E., Bernstein, (2001). P. A., “A survey of

approaches to automatic schema matching”, The

VLDB Journal, 10, pp. 334–350.

Sabou, M., d’Aquin M., Motta, E., (2006). “Using the

Semantic Web as Background Knowledge for

Ontology Mapping”, Proceedings of the 3rd

international conference on Knowledge capture,

ACM, New York, USA, 2006, pp.175-176.

Silva, N., Rocha, J., (2003). “Service-Oriented Ontology

Mapping System”, Proceedings of the Workshop on

Semantic Integration of the International Semantic

Web Conference, Sanibel Island, FL, USA.

Staab, S., Studer, R., (2004). Handbook on Ontologies,

Springer.

STAR, Standards for Technology in Automotive Retail,

(2009). http://www.starstandard.org/.

SUMO, Suggested Upper Merged Ontology, (2009).

http://www.ontologyportal.org/.

WEBIST 2010 - 6th International Conference on Web Information Systems and Technologies