CATCHING CONCEPTS FROM DATABASES BY SCHEMA-BASED COLUMN MATCHING AND CLUSTERING

摘要

To achieve semantic level integration of enterprise databases, it need construct shared concepts hierarchy or ontology for heterogeneous databases to enable semantic interoperability. In terms of database integration, draft concepts can be directly captured by processing schemas of databases. In this context, we propose an automatic approach to capture concepts from relative databases as draft semantic material for ontology, which first carries through matching among relational schema columns based on column names, and then clusters these relational schema columns by joint consideration of schema specification and matching results of column names in a neural network clustering algorithm. In particular, unlike each constraint of schema specification, which is taken as only one classifier for clustering algorithm, the matching results of column names are used as multiple classifiers. So, the whole effectiveness of the approach is not a linear superposition of effectiveness of matching and clustering for each criterion. This is the novelty of our approach.