Esperonto Services IST-2001-34373
Esperonto Services
IST-2001-34373
Deliverable
D14 v1.0
Ontology Alignment Solution
Jos de Bruijn
Ying Ding
Sinuhé Arroyo
Holger Lausen
Institute of Computer Science
University of Innsbruck
{jos.de-bruijn, ying.ding, sinuhe.arroyo, holger.lausen} @uibk.ac.at
10-09-2003
Executive Summary
This deliverable evaluates existing methods and tools for ontology mapping and alignment, defines requirements for an ontology mapping and aligning solutions, and proposes a solution to be used for the SemASP architecture in the Esperonto project.
This is version 1 of the deliverable on Ontology Alignment. The second (and final) version will contain an evaluation of the solution proposed in this version and an update on other ontology alignment methods and tools as their development progresses.
Document Information
Full title / Application Service Provision of Semantic Annotation, Aggregation, Indexing and Routing of Textual, Multimedia, and Multilingual Web Content
Project URL / www.esperonto.net
Document URL
EU Project officer / Brian Macklin
Deliverable / Number / D1.4 / Name / Ontology Alignment Solution v1.0
Task / Number / 1.4 / Name / Ontology Alignment Solution
Work package / Number / 1
Date of delivery / Contractual / 30-08-2003 / Actual / 3011-109-2003
Code name / Status draft ¨ final þ
Nature / Prototype p Report þ Specification p Tool p Other
Distribution Type / Public p Restricted þ Consortium
Authors (Partner) / Jos de Bruijn (IFI), Ying Ding (IFI), Sinuhé Arroyo (IFI), Holger Lausen (IFI)
Contact Person / Sinuhé Arroyo
Email / / Phone / +43 512 507 6488 / Fax / +43 512 507 9872
Abstract
(for dissemination) / This deliverable evaluates existing methods and tools for ontology mapping and alignment, defines requirements for an ontology mapping and aligning solutions, and proposes a solution to be used for the the Esperonto project.
Keywords / Ontology alignment, ontology mapping, information integration
Version log/Date / Change / Author
Project Information
Intelligent Software Components S.A.
(Coordinator) / iSOCO
/ Dr. V. Richard Benjamins
c/ Francisca Delgado 11, 2nd floor
28108 Madrid (Alcobendas), Spain
#e
#t +34-91-334-97-97, #f +34-91-334-97-99
Universidad Politécnica de Madrid / UPM
/ Dr. Asunción Gómez-Pérez
Campus de Montegancedo, sn
Boadilla del Monte, 28660, Spain
#e
#t +34-91 336-7439, #f +34-91 352-4819
Institut für Informatik, Leopold-Franzens Universität Innsbruck / IFI
/ Sinuhé Arroyo
Institute of computer science
University of Innsbruck
Technikerstr. 13
A-6020 Innsbruck, Austria
#e
#t +43 512 507 6485 / 6488
Universität des Saarlandes / UdS
/ Thierry Declerck
DFKI GmbH (German Research Center for AI),
Stuhlsatzenhausweg 3, D-66123 Saarbruecken (Germany)
#e:
#t: +49-681-302-5358, #f: +49-681-302-5338
The University of Liverpool / UniLiv
/ Dr. Valentina A.M. Tamma
Department of Computer Science,
University of Liverpool
Room 1.11, Chadwick Building
Peach Street
Liverpool L69 7ZF, UK
#e
#t +44 151 794 6797, #f +44 151 794 3715
Fundación Residencia de Estudiantes / Residencia
/ Mr Carlos Wert
Fundación Residencia de Estudiantes
Pinar, 23
28006 Madrid, Spain
#e
#t +34-91-446 01 97, #f +34-91-4468068
Centré d'Innovació i Desenvolupament Empreserial / CIDEM
(Centré d'Innovació i Desenvolupament Empreserial)
/ Carlos Gómara
Centré d'Innovació i Desenvolupament Empreserial
Provença, 339
08037 Barcelona, Spain
#e
#t +34-93-4767305, #f +34-93-4767303
Biovista / Biovista
/ Dr. Andreas Persidis
34 Rodopoleos Street
Ellinikon
Athens 16777, HELLAS
#e
#t +30.1.9629848, #f +30.1.9647606
Table of Contents
1. INTRODUCTION 1
1.1. Ontology Alginment in the Esperonto project 2
2. SURVEY OF ONTOLOGY MAPPING & ALIGNING METHODS AND TOOLS 4
2.1. InfoSleuth’s reference ontology 5
2.2. Stanford’s ontology algebra and ONION 7
2.3. AIFB’s formal concept analysis and FCA-Merge 9
2.4. KRAFT’s Ontology Clustering 10
2.5. Chimæra 12
2.6. PROMPT 13
2.7. OBSERVER 15
2.8. Other methods for ontology mapping and aligning 16
2.9. Summary 18
3. A SOLUTION FOR ONTOLOGY MAPPING AND ALIGNING 20
3.1. Problems in Ontology mapping and aligning 21
3.2. Requirements Analysis 22
3.2.1. Mapping Language Requirements 22
3.2.2. Possibilities in automating the creation of mappings 23
3.2.3. User Interface Requirements 23
3.3. Reusing existing methods and tools 24
3.4. Defining the solution 25
3.4.4. The ontology mapping algorithm 25
3.4.5. Integration of the Ontology Alignment module in SemASP 26
4. Conclusions and future work 27
References 28
D1.4. v1: Ontology Alignment Solution V1.0 - ii -
Esperonto Services IST-2001-34373
1. INTRODUCTION
Effective use or reuse of knowledge is essential. Especially nowadays this is the case due to the overwhelming amount of information that is continually being generated, which in turn has forced organizations, businesses and people to manage their knowledge more effectively and efficiently. Simply combining knowledge from distinct domains poses several problems, for instance, different knowledge representation formats, semantic inconsistencies, and so on. The same applies to the area of ontology engineering.
In order to solve these problems the concepts of ontology merging and ontology alignment have been introduced. Noy and Musen (1999) clarified the difference between ontology alignment and ontology merging and noted that “in ontology merging, a single ontology is created, which is a merged version of the original ontologies, while in ontology alignment the two original ontologies remain, with links established between them”. In this report we will refer to these original ontologies as the source ontologies. We will not stress the difference between ontology mapping and ontology aligning. In this report, we deem them as equal[1].
In the Esperonto project, we shall only consider Ontology Aligning and not Ontology Merging. We want to support a distributed architecture with different interconnected ontologies, (possibly) maintained by various different organizations. If we consider ontology merging, the (distributed) source ontologies do not remain and the ontology infrastructure would shift from a distributed to a centralized infrastructure, which is undesirable for our architecture. The ontology maintenance task would shift to one specific organization, so that organizations cannot in general maintain their own ontologies, which would be undesirable from a usability point-of-view and would hinder ontology evolution.
Links (mappings) between concepts in the source ontologies can be represented as conditional rules (Chang & Garcia-Molina, 1998), functions (Chawathe et. al., 1994), logic (Guha, 1991), or a set of tables and procedures (Weinstein & Birmingham, 1998), and so on.
Figure 1
Figure 1 shows a very simple example of ontology mapping between an Employee ontology and a Personnel ontology from different departments of the same company. A different UnitOfMeasure exists in these two ontologies so that the mapping rule of UnitConversion is needed to provide the right mapping between the ‘weight’ properties.
Figure 1: Simple example of ontology mapping (aligning)
1.1. Ontology Alginment in the Esperonto project
The overall objective of the Esperonto project is to provide a bridge between the current web and the Semantic Web. In order to provide such a bridge, the first objective of the Esperonto project is to construct a service that provides content providers with tools and techniques to publish their (existing and new) content on the SW, independently of their native language. This service to be developed is called the SEMantic Annotation Service Provider (SemASP)[2].
Content on the Semantic Web is annotated using ontologies. Because of the distributed nature of the Web, many different providers provide similar content, for example many book vendors will publish their book catalogues on the Web. These different content providers will use different ontologies to annotate their content, since it is hard to agree on a common vocabulary for a large, especially distributed group (Uschold, 2000). The example of different book vendors, annotating their content using (different) ontologies is illustrated in Figure 2.
Figure 2: Example annotation using ontologies
In the example, a user agent that only knows about ontology1 or ontology2, or possibly even only about some other ontology3, will never be able to understand all published catalogues. In order to enable inter-operation between these different representations, there should be a mapping between the different ontologies. When in our example, there exists a mapping between the user agent’s ontology ontology3 and ontology2, the user agent would be able to understand book catalogues B and C. When there would also be a mapping between ontology1 and ontology2[3], then the user agent would also be able to understand book catalogue A.
Figure 3: Example annotation using ontologies with mappings (depicted using straight arrows) in place
Figure 3 shows the mappings between different ontologies and how they enable the user agent to use the different book catalogues.
To summarize, the challenge in ontology alignment in the Esperonto project is to provide explicit mappings between different ontologies, in order to enable inter-operation between different entities on the Semantic Web. In this report we aim to provide a solution for the ontology alignment problem in the Esperonto project, to be implemented as a module in the SemASP architecture.
This report is organized as follows: in chapter 2, we conduct a survey on existing ontology mapping and aligning methods and tools and we provide a summary of the functionalities of the methods. Then, in chapter 3, we will describe our proposed solution for ontology aligning in the Esperonto project. We will describe the requirements we have identified, along with a look into the reuse of existing tools and a proposal for a solution to the ontology alignment problem. Finally, we present some conclusions and an outline of future work.
2. SURVEY OF ONTOLOGY MAPPING & ALIGNING METHODS AND TOOLS
In order to provide a solution for ontology alignment in the Esperonto project, we first conduct a survey on existing methods and tools. One of the goals of this survey is to reuse (elements of) these methods for our solution in the project.
At present there are two main approaches to ontology alignment:
· In the local model, or local ontology, approach the user is represented by an agent in the system and this agent presents the user with its own local data model. The agent performs the translation between the user's local model and either the global model or other local models in order to allow interaction with multiple data sources in the system. And example of the local model approach is the KRAFT project (Preece et al., 2001).
· In the global model, or global ontology, approach the user will view the system through the global data model using a mediator, which is ``a system that supports an integrated view over multiple information sources'' (Hull, 1997). Note that in the local model approach, a user agent will in most cases also contact a mediator in order to allow inter-operation with the system, which contains multiple information sources. An example is the InfoSleuth (Fowler et al., 1999) architecture, where user agents view the individual data models through shared ontologies.
This categorization concerns the run-time approach of ontology mapping, that is, the way translations between different representations are carried out during operation of the system. Another way of categorizing ontology aligning methods is by the approach taken in the creation of the actual mapping during design-time. There are three different types of approaches (i.e. types of mapping patterns) in creating mappings between ontologies (Ding and Foo, 2002):
· One-to-one mapping of ontologies. Mappings are created between pairs of ontologies. Problems with this approach arise when the number of interrelated ontologies is large, which is often the case in organizations where many different applications are in use. The complexity of the ontology mapping problem for the one-to-one approach is O(n2) where n is the number of ontologies. An example of the one-to-one approach is OBSERVER (Mena et al., 2000).
· Using a single-shared ontology. All ontologies in the domain are merged into one ontology. Drawbacks of using a single-shared ontology are similar to those of using any standard (Visser and Cui, 1998). For example, it is hard to reach a consensus on a standard shared among many people (it is always a lengthy process), who typically use a number of different terminologies for the same domain; furthermore, a standard impedes changes in an organization (evolution of standards suffers from the same problems as the development of standards). An example of the single-shared ontology approach is the early stages of the KRAFT project (Visser et al., 1999).
· Ontology clustering based based on the similarity of concepts known to different agents (Visser and Tamma, 1999). The ontology clusters are organized in a hierarchical fashion, where the root node is the most general cluster. A lower level in the hierarchy corresponds to a more agent-specific, less abstract representation of the domain. An example of ontology clustering is illustrated in Figure 4. The agent ontologies are typically mapped to leaves in the tree.
Figure 4: An example of ontology clustering
In the following sections we provide an overview of the state of development of ontology mapping in several projects in the area of ontology mapping and aligning. Information about each project along with the important features associated with the project are provided and highlighted. We will evaluate the described ontology mapping and aligning methods and tools along the two dimensions identified above, distinguishing the run-time and the design-time approach used in each project.
Besides ontology mapping and aligning methods, we have also looked into methods and tools for ontology merging. Ontology merging is similar to ontology aligning, in the sense that during the ontology merging process the same relationships between the concepts in the source ontologies need to be discovered as during the alignment process. In other words, the algorithm for discovering related concepts used in an ontology merging method, can also be used for ontology alignment, and vice versa. The main difference between merging and aligning is the way in which the relationships between concepts are eventually expressed. In ontology merging, concepts based on concepts in the source ontologies are created in a new target ontology, unifying concepts from the source ontologies. In ontology alignment, a mapping is created between similar concepts in the source ontologies, retaining the original ontologies (Noy and Musen, 1999).
A pure ontology merging method or tool does not fit into the categorization for ontology aligning methods we have identified above. The distinction of run-time approaches for ontology alignment, the agent-based and the mediator-based approach, is not applicable to ontology merging, because when ontologies are merged and the source ontologies disappear, no run-time transformations are required. Ontology merging does have some resemblance with the mapping pattern of a single-shared ontology. When ontologies have been merged, the merged ontology can be seen as the shared ontology for the applications that were using the source ontologies.