Research On Graph Partitioning-based Large-ontologies Partitioning And Mapping

Ontology mapping is the key technology of solving the bottlenecks of semantic web development. However, with the developing of semantic web, the large-scale ontologies which have a lot number of concepts and complex relationship between concepts have appeared. Since there are some difference on entity number and mapping difficulty between large-scale ontologies and general ontologies, we should use different mapping method to deal with them. This thesis will focus on the issue of large-scale ontologies mapping.Firstly, the research background of the thesis is briefly introduced, after which the state of the art of the ontology mapping technology is elaborated, as well as the development trend of the mapping technology.Secondly, aiming at the problem of current semantic similarity metric of a single ontology doesn't make full use of the semantic information of ontologies, a new semantic similarity metric based on the feature set of concepts is proposed. It first expresses each concept as a set of features according to the hierarchy of each concept in ontology, and introduces a width influencing factor as the coefficient of each feature. Then, it obtains the concept similarity through calculating the similarity between two sets. At last, we introduce a depth influencing factor, and amend the semantic metric to a more understandable form. Theoretical analysis and experimental results show that the metric is simple, but the results close to human judgment.Thirdly, aiming at the problem of low degree of automation and not uniform in block size for the current large-scale ontology mapping, a new method for large-scale ontology partition and mapping method based on graph partitioning is proposed. It first converts the two ontologies to be matched to DAG structures by preprocessing, which convert the ontologies partition problem into graph partitioning problem, and then partition the two ontologies graphs separately to a set of blocks by using the GPO algorithm which based on genetic algorithm. At last, blocks from different ontology are matched by combining two methods of ontology blocks structure as well as predefined anchors.Finally, LSOPM system has been designed and implemented according to the works above, and compared with the current large-scale ontology mapping system. Experimental results show that this system has a good quality of partition and block mapping and an obvious improvement on both precision and recall.