AN ONTOLOGY MAPPING ARCHITECTURE TO FACILITATE

SEMANTIC INTEGRATION

Yingzi Wang

1,2

, Deren Chen

College of Computer Science, Zhejiang University, Hangzhou, 310027, P. R .China

College of Information Technology, Zhejiang University of Technology, Hangzhou, 310032, P. R. China

Yanyun Wang

Hangzhou Teachers College, 310036, P. R. China

Keywords: Ontology mapping, mediator-centric, semantic integration.

Abstract: Mapping between ontologies is the major bottleneck in semantic integration. In this paper, we present

MCMA, a mediator-centric mapping architecture that facilitates the integration between ontologies of

interrelated companies. In MCMA, different ontologies are mapped through some middle concepts and the

Mapping Service acts as a bridge to connect them. As a mediator, the provider of Mapping Service is

interested in a specific area and offers some common ontologies for participators to reuse. On the

assumption that each individual ontology may reuse or make reference to common definitions, some

mapping relations can be defined conveniently. Based on these basic mapping relations, the Mapping

Service infers more middle-concept oriented mapping relations and organizes them in a layered manner. We

present the idea of combining heuristics or machine-learning techniques with common ontology approach in

the mapping discovery.

1 INTRODUCTION

Heterogeneity is the main problem in the integration

of information systems. Diversified technologies

have been suggested for improving the inter-

operability between heterogeneous systems. Onto-

logies, which capture the semantics of information

from various sources and giving them a concise,

uniform and declarative description (Fensel, 2001),

can make the data more sharable and machine-

understandable. Some researches show that

ontologies are the most likely candidates for solving

the interoperability problems, and there have been

works (e.g. cui, 2002) of using ontologies to

facilitate B2B integration.

Though there are many works in developing

common ontologies, the pervasive adoption of

common ontologies seems unlikely

(Hameed, 2004).

When various ontologies are developed, different

representations and terminologies for the same

concepts emerge and new interoperability problems

come up (Su, 2004). In order to reconcile different

ontologies, it is necessary to establish the mapping

between ontologies.

The ontology mapping is well studied in recent

years. In terms of the mapping architecture, the

common ontology approach or its variant (Hameed,

2004) seems appropriate for some applications, such

as B2B integration. However, it is questionable that

a cluster of individuals would completely agree to

one or several common ontologies. In fact, the more

potential vision is that business partners may reuse

some definitions of common ontologies to develop

their ontologies, and they can extend the common

ontologies with concepts and properties specific to

their applications. In such a scenario, the individual

ontologies don’t absolutely conform to the common

ontologies, thus the common ontologies don’t have

sufficient power to map all the individual ontologies,

and parts of the individual ontologies should be

mapped directly. Unfortunately, the individual

ontologies usually don’t know each other. They need

some support in the mapping discovery and process.

In this paper, a Mediator-Centric Mapping

Architecture (MCMA) is introduced. The initial

134

Wang Y., Chen D. and Wang Y. (2007).

AN ONTOLOGY MAPPING ARCHITECTURE TO FACILITATE SEMANTIC INTEGRATION.

In Proceedings of the Third International Conference on Web Information Systems and Technologies - Internet Technology, pages 134-140

DOI: 10.5220/0001264301340140

 SciTePress

purpose of MCMA is to facilitate the integration of

some inter-related companies, such as the companies

related to certain industry. In MCMA, there is a

Mapping Service acting as a mediator to reconcile

different ontologies. It is based on the following

premise: The provider of Mapping Service is

interested in a specific area and offers some

common ontologies for participators to reuse. The

common ontologies here are just some reference

modules capturing shared understandings of the

specific area. They don’t act as centralized standards

and don’t restrict the participators’ information

descriptions. The participators can make some

extension or modification according to existing data

source or particular business knowledge while

developing their individual ontologies..

On the assumption that each individual ontology

may reuse common definitions more or less, there

are some obvious mappings between individual

ontologies and the common ontologies. And it seems

feasible for the Mapping Service to infer more direct

mappings between individual ontologies. In order to

keep the middle-concept oriented mapping mode,

the inferred direct mappings are transformed to

middle-concept oriented relations.

Content of this paper is structured as following.

Section 2 surveys some related works and provides a

background to our research. Section 3 presents

MCMA, the mediator-centric mapping architecture.

Section 4 describes how the Mapping Service infers

direct mappings between ontologies and transforms

them to virtual mapping relations. Section 5

discusses some features of MCMA. Finally, the

paper ends with a conclusion.

2 BACKGROUND

Mapping between ontologies is the major bottleneck

in semantic integration. In a decentralized view,

every business partner may adopt different

ontologies to represent domain knowledge. Building

direct mapping for each pair of ontologies is

cumbersome or even impossible. Silva argues that

common ontology approach

(Silva, 2002) can reduce

the number of mappings. Using this approach,

individual ontologies are mapped through a common

ontology. Thus they are relatively independent of

each other. In the work of Hameed (Hameed, 2004),

a more manageable, scalable variant of common

ontology approach is introduced. In the variant, each

individual ontology maps to the common ontology

for its cluster, and the common ontologies are

mapped to allow the exchange of information and

knowledge between the clusters.

Though the common ontology approach and its

variant seem potential, we usually do not have the

luck that all ontologies can be mapped through

common ontologies since reaching a consensus is

not easy. If parts of ontologies can’t be mapped

through common ontologies, they should be mapped

directly. Usually, heuristics or machine-learning

techniques are used in the process of inferring direct

mappings between ontologies. For example, GLUE

(Doan, 2002)

applies multiple learners to exploit

information in concept instances and taxonomic

structure of ontologies. It uses a probabilistic model

to combine the results of learners. In another

probabilistic framework (Pan, 2005), ontologies are

firstly translated into Bayesian networks, and the

concept mapping is realized as evidential reasoning

between the two BNs by Jeffrey’s rule.

There are also some researches that try to

develop methods for improving the quality of

existing mappings. OMEN (Mitra, 2005) is an

example that improving existing ontology matches

based on a probabilistic inference. OMEN uses a

Bayesian Net to represent the influences between

potential concept mappings across ontologies, and

uses the mapping to infer mappings between related

concepts. This mapping strategy is also described as

“Taxonomy context based strategy” (Tang, 2005).

Our work is enlightened by above research

works. We are concentrating to design an ontology-

based integration framework for some interrelated

companies. Since these companies are related to a

specific industry or domain, it seems that the

ontology reusing (Ding, 2002) is an applicable

mechanism to simplify the ontology developing and

improve the similarity between different ontologies.

In our integration framework (Wang, 2005), some

common ontology modules are provided for

registered companies to reuse. Due to the reusing,

parts of individual ontologies can be mapped

through common ontologies. However, since the

particular business context can be involved, there are

some concepts that have no correspondence in the

common ontologies existing in various individual

ontologies. Motivated by the analysis above, we

design MCMA, an ontology mapping architecture

that combines Heuristics and Machine-learning

techniques with common ontology approach to

discovery mappings. In our work, some related

researches are adopted and extended to meet the

application scenario.

AN ONTOLOGY MAPPING ARCHITECTURE TO FACILITATE SEMANTIC INTEGRATION

135

3 MEDIATOR-CENTRIC

MAPPING ARCHITECTURE

Figure 1 shows the overview of Mediator-Centric

Mapping Architecture (MCMA). In MCMA, there is

a central Mapping Service that serves as a mediator

to facilitate the mappings between individual

ontologies. Different ontologies are mapped through

some middle concepts and there are mapping

relations that relate concepts in individual ontologies

to middle concepts. The middle concepts are not

limited to the concepts that are described in the

common ontologies, they can be virtual concepts

defined by the Mapping Service. The Mapping

Service is responsible for building and updating the

mapping relations, as well as bridging the individual

ontologies according to the mapping relations.

3.1 Mapping through Middle Concepts

Different ontologies in MCMA are mapped through

middle concepts. The mapping relations serve as the

main evidence for the Mapping Mediation Service to

find the matching pairs between individual

ontologies. A mapping relation is defined as:

MR{O_ID, SC, TC, P, TR

, TR

}

z O_ID identifies an individual ontology.

z SC is a concept in the ontology O_ID.

z TC is the corresponding concept of SC, it may

be either a real concept that is defined in a

common ontology module, or just a virtual

concept that is defined to link related concepts.

z The value of P (from 0 to 1) implies the

possibility of the matching.

z TR

represents the transformation function that

converts SC to TC, while TR

represents the

transformation function that converts TC to SC.

and TR

can be omitted. It means that there

is no transformation function needed or no

transformation function available.

For instance, MR {O

, product_price,

price_USD, 1} means product_price in O

is just the

same as Price_USD in common ontology modules.

In the case such as data integration, the Mapping

Service can use the mapping relations to build the

mappings between ontologies. According to the

mapping relations with the middle concepts, similar

concepts in different ontologies could be mapped to

each other. For example, if there are MR {O

, c

, c,

1} and MR {O

, c

, c, 1} found, the concept c

in the

ontology O

can be mapped to the concept c

in the

ontology O

, and vice versa.

3.2 Layered Mapping Relations

In MCMA, some mapping relations are defined or

validated by human intervention, while some are

inferred by machine. Layered mapping relations

reflect the depth of mapping inference and the

certainty of the mapping relations.

As mentioned previously, MCMA is based on

the premise that most individual ontologies are inter-

related and each ontology makes reference to

common ontologies more or less. So it is convenient

for the participators to define some obvious mapping

relations while developing their ontologies. These

mapping relations can be built in a semi-automated

fashion and submitted to the Mapping Service to

serve as the basic mapping relations, the inner layer

of the layered structure.

The basic mapping relations are not enough in

the process of bridging individual ontologies. There

are two reasons: (1) the basic relations are defined

by human intervention, so it is probable that some

mappings between individual ontologies and

common ontologies are ignored. (2) In the individual

ontologies, there exist some concepts that have no

obvious correspondences in common ontologies.

Basic

Extended

Validated

Virtual

Figure 2: Layered Structure of Mapping Relations.

Figure 1: Mediator-Centric Mapping Architecture.

Identifiers

Common ontologies

Middle Concepts

Layered Mapping Relations

Mapping Service

……

WEBIST 2007 - International Conference on Web Information Systems and Technologies

136

These concepts can’t be mapped through common

ontologies. Therefore, in order to reconcile different

ontologies, the Mapping Service should do the

mapping inference. Since there have been basic

mapping relations, taxonomy context based strategy

(Tang, 2005) seems appropriate.

Mapping inference in MCMA consists of two

phases. At first, the Mapping Service infers more

mapping relations between individual ontologies and

common ontologies on the basis of basic mapping

relations. The inference task is disassembled to a

series of sub-tasks. A sub-task is responsible for

inferring more mappings between two ontologies, a

common ontology and an individual ontology that

have been partly mapped. A sub-task is done by a

relatively independent sub-service of the Mapping

Service, Direct Mapping Enhancer. The mapping

relations infered from this phase are termed as

extended mapping relations, which locate in the

middle layer of the layered structure. In this layer,

the validated relations are distinguished from the

other relations. They represent some of the extended

relations that are validated by the owners of

participated ontologies.

Secondly, According to the previous two layers

of mapping relations, the Mapping Service infers

more mappings between individual ontologies. In

order to be consistent with the middle-concept

oriented mapping mode, the inferred mappings are

transformed to virtual mapping relations, in which

concepts are mapped through virtual concepts, some

identifiers defined by the Mapping Service. The

virtual mapping relations form the outer layer of the

layered structure. The inference of virtual relations

is more difficult, and we will discuss the inference

of virtual mapping relations in section 4.

3.3 The Management of Mapping

Relations

The Mapping Service and related participators

operate in a highly dynamic environment. It is

improper to regard the MCMA as a static

architecture. New ontology may join, or an existing

ontology may drop out. Furthermore, all the

ontologies, including individual ontologies and

common ontologies, may evolve constantly. Thus

the Mapping Service should manage the change of

the Mapping Relations.

In the layered structure of mapping relations, a

layer depends on the layers under it. Accordingly the

change of a layer may cause the modification of the

layer above it. For example, if the basic mapping

relations are changed, the Mapping Service should

rebuild the extended mapping relations and the

virtual mapping relations.

Most changes are introduced by the variation of

ontologies. When a participator’s ontology is

modified, the participator should propagate the

changes to the Mapping Service and the Mapping

Service should revise the mapping relations that

involve the changed ontology. The situation is more

complex when changes occur in a common

ontology. The Mapping Service should notify all the

participators who make reference to the changed

common ontology and the participators should make

modification to the basic relations that involve their

ontologies. If some participators couldn’t do the

modification, the Mapping Service would delete the

relations that seem doubtful due to the variation of

common ontologies. Obviously, the modification of

common ontologies can cause the updating work

very hard, so it is recommended that only stable

concepts be defined in the common ontologies.

4 BUILDING VIRTUAL MAPPING

RELATIONS

The virtual mapping relations are inferred on the

basis of the basic mapping relations and the

extended mapping relations. The major steps are

described as follows: Firstly, according to the

existing mapping relations, the Mapping Service

builds primary mappings for each pair of ontologies.

Secondly, according to the primary mappings and

the semantic relations between concepts, the

Mapping Service infers more direct mapping

relations between individual ontologies. Finally, the

inferred relations are transformed to the virtual

relations that are middle-concepts oriented. In this

section, we will discuss how new mappings are

inferred and transformed to the virtual relations. In

order to reduce the complexity of mapping

inference, we ignore the transformation between

concepts while doing the inference. So a mapping

relation is denoted as MR {O_ID, SC, TC, P} in the

inference.

4.1 Direct Mapping Inference

According to the basic mapping relations and the

extended mapping relations, the Mapping Service

can build some primary mappings for each pair of

individual ontologies. Once the Mapping Service has

built the primary mappings for each pair of

ontologies, the next step is to infer more direct

mappings between them.

Except the primary mapping relations between

ontologies, the semantic relations between concepts

AN ONTOLOGY MAPPING ARCHITECTURE TO FACILITATE SEMANTIC INTEGRATION

137

are used to infer new mapping relations. A semantic

relation is defined as:

SR {O_ID, C1, C2, R}

z O_ID identifies an ontology. If the O_ID is

omitted, it means that the semantic relation is

about the concepts in common ontologies.

z Both C1 and C2 are concepts in the ontology

O_ID.

z R refers to the semantic relationship between

C1 and C2, such as Is_a, Part_of,

Disjoint_with and Overlap_to.

Therefore, based on the primary mapping

relations and semantic mapping relations, the Direct

Mapping Enhancer infers more direct mappings

between two ontologies. These new mappings

cluster to form the set of Direct Mappings (DM),

which serve as the key evidence to build the virtual

mapping relations.

Threshold is given to decide whether the direct

mapping inference for a pair of ontologies can be

carried out. Only if the primary mappings between

two ontologies are above the threshold (i.e. more

than 20% of the concepts have been mapped), the

Mapping Service does the mapping inference for

them. Once a pair of ontologies has passed the infer-

ability examination, a sub-service, the so-called

“Direct Mapping Enhancer” is invoked to infer more

mapping relations between them. Figure 3 shows the

mapping inference algorithm.

In MCMA, the Direct Mapping Enhancer is a

relatively independent sub-service that infers more

mappings between two partly mapped ontologies. In

this paper, we mainly concern the middle-concept

oriented mapping architecture. So we don’t discuss

the algorithm of Direct Mapping Enhancer in detail.

In principle, it can be any related techniques adopt

from the current and future researches, and it can be

updated whenever necessary. Currently we mainly

adopt some ideas from OMEN (Mitra, 2005). And

we find that the result of mapping inference is ideal

if two ontologies have the similar structure.

However, if there is structure discrepancy between

two ontologies, the inference work is hard and the

result is still not satisfactory enough. We are trying

to improve the algorithm of Direct Mapping

Enhancer to get more ideal mapping result.

4.2 Transforming Direct Mappings to

Virtual Mapping Relations

Once the Mapping Service has inferred more direct

mappings between ontologies, the next step is to

generate new mapping relations according to the

inferred direct mappings. In order to be consistent

with the middle-concept oriented mapping

mechanism of MCMA, the Mapping Service should

transform the inferred direct mappings to the virtual

mapping relations. Some identifiers are introduced

to serve as the middle concepts when building the

virtual mapping relations. The following

summarizes the transformation algorithm:

Input: DM, the set of inferred direct mappings.

Output: VMR, the set of virtual mapping relations.

Step:(1) In terms of DM, calculate the total-

probability of mappings for every concept that

appears in DM. Let P={ p(O

, c

), p(O

, c

)…p(O

)…}be a set of total-probability, and p(O

, c

)

represents the total-probability of the concept c

ontology O

. For each { O

, c, O

, c’, p}∈DM, do

p(O

, c)= p(O

, c) +p and p(O

, c’)= p( O

, c’)+p.

(2) Select the maximum value p(O

max

, c

max

) from

P. Then a new identifier id is defined to replace the

However, the Mapping Service does not infer

mappings for every pair of ontologies. Before

Input:

PM={MR

,MR

,….,MR

represents an existing mapping relation.

SR={ SR

, SR

, ….,SR

…., SR

}

represents the set of the semantic relations of O

Step:

For each middle concept c that appears in PM do

While there are {O

, c

c, p

}∈PM

and {O

, c

, c, p

}∈PM do

Add concept pair (c

, c

, p

) → M

i,j

Delete {O

, c

c, p

} and {O

, c

, c, p

} from PM

End While

End For

For i=1,….n-1 do

For j=i+1 ….n do

the number of concepts pairs in M

i,j

→ num1

the maximum number of concepts

in O

→ num2

If num1/num2>= threshold then

(1) Invoke “direct mapping enhancer”, infer

more mappings between O

on the basis

of M

i,j

, SR

and SR

;

(2) The set of new mappings between

→NM

i,j

(3) For each (c

, c

, p)∈NM

i,j

If p> a given threshold Then

Add { O

, c

, p}→ DM

End For

End If

End For

Output:

DM, the set of new direct mappings

Figure 3: The Mapping Inference Algorithm.

WEBIST 2007 - International Conference on Web Information Systems and Technologies

138

inferring new mappings, the Mapping Service exams

the infer-ability for each pair of ontologies. A

concept c

max

in O

max

. Add { O

max

, c

max

, id,1}to VMR

and invoke the recursive function FindSimilar(O

max

max,

id) to find concepts that is similar to c

max

and do

the transformation work.

(3) Do (2) until DM=ф

The total-probability of a concept indicates how

probably the concept maps to the other concepts in

the DM. It seems suitable to select a concept with

larger total-probability to act as the middle concept.

The main purpose of the Function FindSimilar is

to find the concepts that can be matched to the

identifier id as much as possible from the direct

mapping set DM. The recursive process restrain the

situation that more than one identifier are created to

replace the completely same concepts which are

defined in different ontologies.

5 DISCUSSIONS

The main purpose of MCMA is to facilitate the

integration of some interrelated companies. Our

early research focuses on the following question:

Does the MCMA seem promising in the application

scenario? To answer this question, we have

investigated some companies that are related with

the ironware industry. As expected, despite the terms

and information models are diversified, there are

still shared understandings among these companies.

Accordingly, though these companies can’t

absolutely conform to one or several common

ontologies, they can benefit from the reusing of

some shared ontologies while developing their

individual ontologies(Wang, 2005).

Therefore, we think it is the more convincible

architecture that companies can freely reuse contents

of common ontologies to develop their individual

ontologies. The common ontologies are shared

modules capturing relevant concepts and knowledge

that most companies expected. They may be stored

in a library. The related companies can select desired

modules on it’s own account and custom-modulate

to meet their needs while developing individual

ontologies. In such a scenario, only parts of

individual ontologies can be mapped through

common ontologies. So the common ontology

approach (Silva, 2002) seems not competent here.

MCMA is designed for the integration scenario

above. In MCMA, the Mapping Service can discover

more potential mappings by combining the mapping

inference technologies with the common ontology

approach. As we have mentioned before, the main

feature of MCMA is mapping ontologies through

middle-concepts. While transforming the inferred

direct mappings to the middle-concept oriented

relations, virtual concepts are introduced to link

concepts between ontologies.

The primary reason of using virtual concepts to

replace the actual concepts is to reduce the

dependency between ontologies. For example,

supposing there are direct mappings {O

,1}, {O

,0.7}, and {O

,0.8}in

DM, and id

is defined to replace concept c

, the

virtual relations MR{O

,id

,1}, MR{O

,id

,1},

MR{O

, c

,id

,0.7} and MR{O

, id

, 0.8} would

be appended to the VRM. Thus the concept c

, c

and c

can be mapped to each other through id

. If

the ontology O

drops out from the mapping

architecture, what needed is deleting MR{O

,1} from the VMR. The concept c

, c

and c

can

still be mapped through id

From the above example, we also notice that

more concepts can be matched with each other after

the direct mappings are transformed to virtual

mapping relations. In the above example, either c

can be mapped to c

according to the direct

mappings in DM, but

there is no direct mapping

relation between c

and c

. In other words, c

and c

can’t

be mapped to each other in the direct mapping

mode. However, while the direct mappings are

transformed to the middle concept oriented mapping

relations, c

and c

can be mapped through the virtual

concept id

In our experiment, we find a mapping problem

that caused by reduplicate and inconsistent virtual

mapping relations. In the previous example, c

and

can be mapped through id

. Supposing that there

is also a direct mapping relation {O

,0.75}

in DM, and two virtual mapping relations

MR{O

,id

,1}and MR{O

, id

, 0.75} are

appended to the VRM during the transformation, c

and c

would be mapped through id

. It seems

FindSimilar (O

rep

, c

rep

, id)

{

while there exist mapping relations that

involve (O

rep

, c

rep

) in DM, do

{ select a mapping relation { O

, c

rep

, c

rep

, p}

or { O

rep

, c

rep

, O

, c

p} from DM;

add virtual relation{ O

, c, id, p} to VMR;

delete the selected relation from DM;

if(p==1) FindSimilar (O

, c, id);

}

Figure 4: The Recursive Function—FindSimilar.

AN ONTOLOGY MAPPING ARCHITECTURE TO FACILITATE SEMANTIC INTEGRATION

139

trouble when the Mapping Service tries to match c

and c

since they can be mapped through id

or id

with different mapping probabilities. We called this

phenomenon “mapping inconsistency”.

The inconsistency would not happen while the

Mapping Service was building the exact mapping for

ontologies. In this situation, only certain mapping

relations (the possibility value p=1) are selected to

build the mappings. However, if the mapping

process is carried out in the situation that precision

of information is not very important, such as

information searching, the mapping inconsistency

may occur since some uncertain mapping relations

would be selected to build the mappings.

Fortunately, in this situation, we mainly concern the

maximum possibility that two concepts would be

matched. Thus inconsistency of the mapping

relations doesn’t impede the Mapping Service to

discover the matching possibility of two concepts.

MCMA seems suitable for our integration

scenario. Since all the individual ontologies in

MCMA are mapped through middle concepts, they

are relatively independent of each other.

Furthermore, The middle-concept oriented approach

makes it convenient to map one concept in certain

ontology to multiple concepts in other ontologies. So

MCMA is especially suitable for the situation that

1:n mapping is necessary, such as information

searching.

Since ontologies may evolve constantly, the

update of mapping relations in MCMA is crucial. In

our work, we assume that the Mapping Service can

get notified if there is any change happening, and it

would rebuild the mapping relations when

necessary. Furthermore, if there are changes

happening in the common ontologies, the changes

can propagate to the related participators, who

would modify the basic mapping relations.

6 CONCLUSIONS

In this paper, we have presented MCMA, an

ontology mapping architecture that facilitates

semantic integration for some inter-related

companies. Obviously, MCMA is enlightened by the

common ontology approach (Silva, 2002). But it is

somewhat different from the usual common

ontology approach. In MCMA, the mapping

relations, that serve as the main evidence for the

Mapping Service to bridge individual ontologies, are

arranged in a layered manner. Thanks to the reusing,

the basic mapping relations, the first layer of the

layered structure, can be defined conveniently by

participators. And mapping relations of other layers

can be inferred on the basis of the basic relations.

From the inference of virtual mapping relations, we

see the probability of combining heuristics or

machine-learning techniques with common ontology

approach in the mapping discovery.

Though MCMA is middle-concept oriented

architecture, the direct mapping inference is a

crucial step in the building of layered relations. In

future, we plan to improve algorithms of the direct

mapping inference. Besides, future work also

includes developing an integrated mechanism of

managing and updating the layered mapping

relations.

REFERENCES

Cui, Z., Jones, D. and O’Brien, P., 2002. Semantic B2B

Integration: Issues in Ontology-based Approaches.

ACM SIGMOD.Vol. 31, issue 1.

Ding, Y., Fensel, D., Klein, M. and Omelayenko, B.,

2002. The semantic web: yet another hip? Data &

Knowledge Engineering Vol. 41.

Doan, A., Madhavan, J., Domingos, P., and Halevy, A.,

2002. Learning to map between ontologies on the

semantic web, Eleventh International WWW

Conference (WWW2002).

Fensel, D., 2001. Ontologies: A Silver Bullet for

Knowledge Management and Electronic Commerce.

Springer.

Hameed, A., Preece, A., Sleeman, D., 2004. Ontology

Recon-ciliation. In Staab, S., Studer, R. (Eds.):

Handbook on Ontologies. Springer.

Mitra, P., Noy, N. F., Jaiswa, A., 2005. OMEN: A

Probabilistic Ontology Mapping Tool. International

Semantic Web Conference 2005.

Pan, R., Ding, Z., Yu, Y., and Y. Peng, 2005. A Bayesian

Network Approach to Ontology Mapping,

International Semantic Web Conference 2005.

Silva, N., Rocha, J., 2002. Ontology Mapping Using

Multiple Dimension Approach, International

Conference on Fuzzy Systems and Soft Computational

Intelligence in Management and Industrial

Engineering.

Su, X., 2004. Improving semantic interoperability through

ana-lysis of model extension. Proc. of 9th

International Con-ference on Applications of Natural

Language to Information Systems (NLDB'04).

Tang, J., Liang, B., and Li, J., 2005 Toward Detecting

Mapping Strategies for Ontology Interoperability,

Workshop on Semantic Computing in the 14

International world wide web coference (WWW2005).

Wang, Y., Chen, D., Zheng, X., 2005. B2B Integration:

An Ontology-Based Framework. Proceedings of 1st

International Conference on Semantics, Knowledge

and Grid (SKG 2005), IEEE Computer Society.

WEBIST 2007 - International Conference on Web Information Systems and Technologies

140