Large-scale Ontology Matching On MapReduce

With the rapid development of Semantic Web, more and more Ontologies were released. The wide use of ontologies brings a practical problem, that is, ontologies coming from overlapped domains contain heterogeneous classes, properties and instancs. Ontology Matching is a critical tool that realizes mappings between different ontologies, to cope with the heterogeneity. However, the traditional Ontology Matching algorithms cannot deal with the large-scale Ontology Matching problem, because most of them failed to guarantee both effectiveness and efficiency. Researchers have tried to solve this problem by simplifying the algorithms and dividing ontologies. The latter one may reduce the computation space but does not break through the limitations of single machine power.In the recent years, MapReduce distributed computing framework has attracted the attention of researchers. MapReduce framework divides computation tasks into various computing node through its key-value combination mechanism. In fact, the large-scale Ontology Matching problem can be divided into 2 categories:large ontology pairwise matching and large-scale multi-ontology matching. Large ontology pairwise matching deals with the ontology matching problem involving 2 ontologies with large size; the second one, different from large ontology pairwise matching, copes with multi-ontology matching. The approach in this paper uses MapReduce to solve large-scale ontology matching problem. The contribution of this paper includes the following points:(1) Large ontology pairwise matching based on MapReduce. This paper presents an ontology matching algorithm based on Virtual Document similarity and MapReduce framework to solve large ontology pairwise matching problem. This method takes advantage of three features of MapReduce framework, that is, data connection, graph traversal and data partitioning. This method, called V-Doc+, can be divided into three MapReduce stages. Firstly, the MapReduce process connected each entity (class, property and instance) and blank node with the subject-related RDF statements, to construct its description. Then, the approach takes 1-step graph traversal in a MapReduce process so that every entity can aggregate with its adjacent nodes, to enrich its local description. Next, the approach uses a word weight-based partition method and distributes potential matched entities into same reducer. Experimental results show that V-Doc+is able to guarantee both effectiveness and low run time.(2) Large-scale multi-ontology matching based on MapReduce. Different from large ontology pairwise matching, large-scale multi-ontology matching needs to take numbers of pairwise matching task among several ontologies. Even a single ontology is small, the big combination of these ontologies makes the pairwise matching process repeats times. Therefore, this paper presents two parallel multi-ontology matching schemes based on MapReduce framework, and make comparative experiments between them. Compared with sequential ontology pairwise matching, these methods lead to a substantial reduction in run time.