A Framework for Efficient Matching of Large-Scale Metadata Models

Despite the success achieved in the metadata models matching area, large-scale matching does not preserve high match quality and efficiency at the same time. To deal with these challenges, we introduce a generic matching framework, called MetMat, to identify and discover corresponding entities across XML schemas and/or ontologies (metadata models). In particular, the proposed framework is based on a parallelized clustering-based matching approach, which first splits the original matching task into smaller independent tasks. These independent tasks are then carried out in parallel exploiting desktop platform features that are equipped with parallelism enabled multi-core processors.  To this end, we develop three different parallel strategies: inter-, intra-, and hybrid-matching strategies. To obtain high quality, a set of matchers are exploited. The proposed framework is validated through an extensive set of experiments over small and large datasets. We also compared the MetMat framework to top matching tools participating in the OAEI (Ontology Alignment Evaluation Initiative)
for the last three years. The results show that the MetMat framework
with the intra-parallel matching strategy outperforms other matching strategies in terms of processing time while preserving the same quality. Moreover, the tool acquires a good position through OAEI for the last
three years.