| Ontology is an explicit specification of concepts, relations and other related elements. Ontologies are often used to represent different domain knowledge, and become the foundation for information integration and system interoperation among different applications. However, the differences among individual ontology designers will result in heterogeneity on proposed domain onotologies. And it may inevitably affect inter-operation between resulting systems.In order to facilitate inter-operation between different applications, it is necessary to build a mapping model for those heterogeneous ontologies, which requires to find mapping relationships between correspondent elements. In recent years, some researchers have proposed a number of mapping approaches, mainly including Cupid, COMA, Glue, Rondo, S-Match, and so on. These approaches can be classified into three categories: terminology approach, structural approach and logic approach. The terminology approach is easier and convenient use, but when there is lacking of the apriority knowledge on concept and terms, it is difficult to achieve a good performance. Structural approaches will not be able to handling ontologies which are heterogeneous in nature or have many logical constraints. The logic approaches based on logical reasoning are often too strict to capture all valid mappings. Another common problem of those methods are efficiency problem, they cannot support the efficiency demands of large scale ontology mapping.This paper focuses on two problems:1. How to further improve the accuracy of ontology mappings?;2. How to handle large scale ontology mapping more effectively?Following are main contributions and achievements:1. An Ontology Parsing graph-based Mapping (OPM) algorithm is proposed. It constructes two parsing graphs for specified ontologies first and then transfer the ontology mapping to a problem of finding the maximal match between the two OP-graphs. Through continuously updating the iteratively similarity of those OP-graph vertices, the maximal match of the two OP-graphs can be calculated. Experiments show that OPM algorithm is applicable to different kinds of ontologies, and have both average precision and recall ratio (0.95, 0.90) 6 percent higher than the Fujitsu algorithm (0.89, 0.84) which is the best among other methods.2. In order to further improve the average precision of OPM, an axiom based ontology revising method (ARevision) is presented. Axioms are often used to describe the intrinsic constraint relationships among ontologies concepts, elements and other related items. By checking violations of these intrinsic relationships, possible false mappings could be identified and eliminated. Experiments show that by using ARevision, the average precision of OPM can be further improved by another 2-3 percents.3. Proposed a Modular Ontology Mapping (MOM) method for large scale ontology mappings and conducted study on how to effectively partition a large scale ontology into a number of small modules. By using Hopcroft-Karp algorithm, we transform a large scale ontology matching problem into a number of smaller scale module matching problems, which greatly reduced the complexity of the problem nature and speed up the process. The Hopcroft-Karp algorithm, however, only applicable to some ontologies which have good modularity in nature. To dealing with more general large scale cases, a cluster based partition method CBpartition is presented. This method has following characteristics:(1) it is suitable for all complex and large scale ontologies;(2) Since the module mapping later involves only the similar parts of the original ontologies, it effectively improves the efficiency of both partition and mapping process.
|
PDF Full Text Request