Ontology Matching based on Combination of Lexical and Structural Techniques in Semantic Web
Minh-Diep Nguyen, Thi Thuy Anh · 2016
Nowadays, ontologies become the foundation of Semantic Web. The number of ontologies is increasing day by day. Researching on ontologies and its applications in various fields such as artificial intelligence, computational linguistics, computer science, e-commerce have been spreading and maturing. Actually, ontologies represent the characteristics of a specific domain. They include classes, properties, relationships, and instances. Since being different from background knowledge, languages used for expression, points of view of designers and entities are modeled in different ways, there is not one ontology matched perfectly to another one. This leads to heterogeneity between ontologies. In addition, the management of knowledge based on ontologies is necessary. Therefore, comparing, mapping, and integrating ontologies should be implemented in which the task of matching is to reduce ontology heterogeneity problem and identify the similarities between entities from ontologies. From the issue mentioned above, research communities have developed methods for ontology matching based on several aspects of similarity such as lexical, semantic, structural, and instances. This thesis focuses on the task of ontology matching which has received many investigations in recent years. Although a lot of individual similarity measures are proposed, no ontology matching system uses only one technique to match. Normally, more than one similarity measure is used and the matching results are then combined to obtain the final alignment. Ontology matching systems give solutions to achieve the best possible matching results by using lexical-based, structure-based, semantic-based, and instances-based techniques together. The proposed methods used in these systems take into account different aspects of the similarity of entities in ontologies. In this work, ontology matching is based on our structural, lexical and semantic methods and use WordNet dictionary. In particular,we present an improvement of the lexical metric by applying information theoretic and edit distance approaches, new structural and semantic measures. The first contribution of this study is applying information-theoretic and edit distance methods to flexibly measure lexical similarity. Besides of improving the accuracy for string-based similarity degrees, this metric deals with some irrelevant situations. Our second approach is a novel structure-based similarity measure for automatic ontology matching. Being different from existing structural measures, this approach takes into account all of the ancestors of considered concepts. Another contribution of this research is a semantic similarity measure between nouns based on the structure of WordNet. This measure uses the WordNet dictionary as an external resource to take semantics of entities. Besides the positions of two entities relatively to the root in a hierarchy, this approach considers the relationships between these entities. Our ontology matching solution is integrated by using weighted sum method to measures in which both sequential and parallel strategies are executed for computing similarity. After that, we will evaluate the quality of our system. Our approach is implemented on the benchmark dataset of the 2008 OAEI and then compared to the other systems. The experimental results show that our approach reaches good F-measure values and can compete with other automatic systems which do not use instances. The one-to-one or one-to-many alignments are generated in the final phase. The approaches presented in this thesis could also be applied in many application domains.