Ontology Alignment (eBook)

Contents 6
List of Figures 12
Preface 16
Introduction and Overview 19
1.1 Motivation 19
1.2 Contribution 21
1.2.1 Problem Outline 21
1.2.2 Solution Pathway 22
1.3 Overview 23
1.3.1 Structure 23
1.3.2 Reader's Guide 24
Part I Foundations 26
Definitions 27
2.1 Ontology 27
2.1.1 Ontology Definition 27
2.1.2 Semantic Web and Web Ontology Language (OWL) 30
2.1.3 Ontology Example 32
2.2 Ontology Alignment 35
2.2.1 Ontology Alignment Definition 35
2.2.2 Ontology Alignment Representation 36
2.2.3 Ontology Alignment Example 37
2.3 Related Terms 39
2.4 Ontology Similarity 41
2.4.1 Ontology Similarity Definition 41
2.4.2 Similarity Layers 42
2.4.3 Specific Similarity Measures 44
2.4.4 Similarity in Related Work 50
2.4.5 Heuristic Definition 50
Scenarios 52
3.1 Use Cases 52
3.1.1 Alignment Discovery 53
3.1.2 Agent Negotiation / Web Service Composition 53
3.1.3 Data Integration 54
3.1.4 Ontology Evolution / Versloning 55
3.1.5 Ontology Merging 55
3.1.6 Query and Answer Rewriting / Mapping 56
3.2 Requirements 57
Related Work 59
4.1 Theory of Alignment 59
4.1.1 Algebraic Approach 59
4.1.2 Information- Flow- based Approach 60
4.1.3 Translation Framework 61
4.2 Existing Alignment Approaches 61
4.2.1 Classification Guidelines for Alignment Approaches 61
4.2.2 Ontology Alignment Approaches 63
4.2.3 Schema Alignment Approaches 67
4.2.4 Global as View / Local as View 70
Part II Ontology Alignment Approach 73
Process 74
5.1 General Process 74
5.2 Alignment Approach 77
5.2.0 Input 77
5.2.1 Feature Engineering 78
5.2.2 Search Step Selection 80
5.2.3 Similarity Computation 81
5.2.5 Interpretation 85
5.2.6 Iteration 87
5.2.7 Output 88
5.3 Process Description of Related Approaches 89
5.3.1 PROMPT, Anchor- PROMPT 89
5.3.2 GLUE 91
5.3.3 OLA 92
5.4 Evaluation of Alignment Approach 94
5.4.1 Evaluation Scenario 94
5.4.2 Evaluation Measures 95
5.4.3 Absolute Quality 101
5.4.4 Data Sets 101
5.4.5 Strategies 104
5.4.6 Results 105
5.4.7 Discussion and Lessons Learned 108
Advanced Methods 110
6.1 Efficiency 110
6.1.1 Challenge 110
6.1.2 Complexity 111
6.1.5 Discussion and Lessons Learned 119
6.2 Machine Learning 120
6.2.1 Challenge 120
6.2.2 Machine Learning for Ontology Alignment 121
6.2.3 Runtime Alignment 126
6.2.4 Explanatory Component of Decision Trees 127
6.2.5 Evaluation Scenarios: Training and Test Data Sets 128
6.2.6 Discussion and Lessons Learned 130
6.3 Active Alignment 132
6.3.1 Challenge 132
6.3.2 Ontology Alignment with User Interaction 133
6.3.3 Evaluation 134
6.4 Adaptive Alignment 137
6.4.1 Challenge 137
6.4.2 Overview 138
6.4.3 Create Utility Function 138
6.4.4 Derive Requirements for Result Dimensions 140
6.4.5 Derive Parameters 141
6.4.6 Example 144
6.4.7 Evaluation 145
6.4.8 Discussion and Lessons Learned 146
6.5 Integrated Approach 148
6.5.1 Integrating the Individual Approaches 148
6.5.2 Summary of Ontology Alignment Approaches 149
6.5.3 Evaluation 149
6.5.4 Discussion and Lessons Learned 151
Part III Implementation and Application 156
Tools 157
7.1 Basic Infrastructure for Ontology Alignment and Mapping - FOAM 157
7.1.1 User Example 157
7.1.2 Process Implementation 158
7.1.3 Underlying Software 159
7.1.4 Availability and Open Usage 160
7.1.5 Summary 161
7.2 Ontology Mapping Based on Axioms 161
7.2.1 Logics and Inferencing 162
7.2.2 Formalization of Similarity Rules as Logical Axioms 163
7.2.3 Evaluation 164
7.3 Integration into Ontology Engineering Platform 165
7.3.1 OntoStudio 165
7.3.2 OntoMap 166
7.3.3 FOAM in OntoMap 167
Semantic Web and Peer-to-Peer — SWAP 168
8.1 Project Description 168
8.1.1 Core Technologies 169
8.2 Bibster 170
8.2.1 Scenario 171
8.2.2 Design 171
8.2.3 Ontology Alignment / Duplicate Detection 174
8.2.4 Application 177
8.3 Xarop 178
8.3.1 Scenario 178
8.3.2 Design 180
8.3.3 Ontology Alignment 184
8.3.4 Application 185
Semantically Enabled Knowledge Technologies - SEKT 186
9.1 Project Description 186
9.2 Intelligent Integrated Decision Support for Legal Professionals 188
9.2.1 Scenario 188
9.2.2 Use Cases 188
9.2.3 Design 189
9.3 Retrieving and Sharing Knowledge in a Digital Library 190
9.3.1 Scenario 190
9.3.2 Use Cases 190
Part IV Towards Next Generation Semantic Alignment 193
Next Steps 194
10.1 Generalization 194
10.1.1 Situation 194
10.1.2 Generalized Process 195
10.1.3 Alignment of Petri Nets 196
10.1.4 Summary 200
10.2 Complex Alignments 201
10.2.1 Situation 201
10.2.2 Types of Complex Alignments 202
10.2.3 Extended Process for Complex Alignments 203
10.2.4 Implementation and Discussion 204
Future 205
11.1 Outlook 205
11.2 Limits for Alignment 207
11.2.1 Errors 207
11.2.2 Points of Mismatch 208
11.2.3 Implications 209
Conclusion 211
12.1 Content Summary 211
12.2 Assessment of Contribution 213
Part V Appendix 217
A Ontologies 218
B Complete Evaluation Results 222
C FOAM Tool Details 228
References 233
Index 250

7 Tools (S. 145-146)

After the creation of the new approaches, we now need to deploy them. Therefore, we will first show how these are implemented. This chapter is split into three implementations of the alignment approach: a basic infrastructure for ontology alignment and mapping (FOAM), Ontology Mapping based on Axioms (OMA), and the OntoStudio plug-in OntoMap. All of them have specific strengths, though the most important implementation undoubtedly is FOAM. We will refer to it in the subsequent practical applications.

7.1 Basic Infrastructure for Ontology Alignment and Mapping - FOAM

Many methods and approaches for ontology alignment have been presented in this work, but a theoretical framework alone is not sufficient. One needs a proof of concept. In the beginning of this work we made another claim: In fact, we want a system suiting practical applications. Therefore, the goal of our Framework for Ontology Alignment and Mapping (FOAM) is twofold: first, an environment for testing of our new methods for alignment, and second, a stable ontology alignment system. The infrastructure fulfilling the two goals will now be described in detail.

7.1.1 User Example
To give an example, our user wants to align the two ontologies animals A. owl and animalsB.owl. The goal is to find the ahgnments. They are supposed to be general enough for any usage afterwards. Therefore, she starts the FOAM tool from command line (Figure 7.1) with the corresponding parameters (Example 7.1), i.e., the two ontologies and the alignment discovery scenario. The whole process takes time, but this is not a critical factor for this use case. After finishing, FOAM saves the results in a file. The results are of high quality with interesting nontrivial alignments, e.g., a two-legged person in one ontology is equivalent to a bipedal person in the other one.

7.1.2 Process Implementation

We will first generally present FOAM with its parameters and the actual implementation of the process.

FOAM provides three modes for execution. It may be accessed through the command line using a simple parameter file. Further, as it has been implemented in Java,^ it may be run through a Java API making it easy to connect to other applications. It has been run successfully on PCs under the Windows and Linux operating systems. And finally, for lightweight testing purposes, it can be run on our server as a web service which can easily be accessed via a web interface.^ According to the twofold goal of FOAM the user can either specify all parameters himself (strategy, number of iterations, semi-automatic or not, etc.) or let the system set the optimal parameters based on the ontologies to align and the desired use case. A NOSCENARIOoption is available, if the use case is unspecified. The parameters are either set in a separate parameter file or directly handed over to the alignment class within the Java application. Example 7.1 contains two ontologies to align, the scenario, and the filename for the results.

The process of ontology alignment has been implemented as follows. FOAM requires two ontologies as input, in particular ontologies in OWLDL. Besides, preknown alignments may be entered into the system. Figure 7.1 shows the command fine call, the output information of FOAM on the used strategy, the process with several iterations, and the location where the results are saved. Alignments are provided either in the RDF-representation format or as a comma-separated list. The search step selection is implemented with difi'erent kinds of agendas. FOAM includes flexible classes for feature selection (RDF(S) and OWL), as well as a hbrary of similarity measures (also using external information such as dictionaries). Afterwards, there are different combination strategies ranging from simple averaging to machine learned decision trees or neural nets. Whereas the iteration step is trivially implemented through a loop, the interpretation allows for human interaction, where the user has to confirm or reject questionable alignments. If required, the one alignment link strategy is enforced during the interpretation step. The final results are, according to the different implementation goals, either a file of alignments above the threshold and doubtable alignments (for human interaction), or a complete file of alignments ranked by their confidence value. If a gold standard is provided, the alignments are automatically evaluated and the evaluation results are stored as well. All methods of the previous chapters have been implemented. Sometimes, they have been extended with additional features, if this made the framework more flexible for future extensions.