CLUSTERING DATABASE OBJECTS FOR SEMANTIC INTEGRATION OF HETEROGENEOUS DATABASES

摘要

Interschema Relationship Identification (IRI), i.e., determining the relationships between objects inrnheterogeneous database schemas, is critical to both the classical schema integration problem and the datarncleansing and consolidation phase that precedes data warehouse development. In this paper we propose arncluster analysis-based approach to semi-automate the IRI process, which is typically very time-consuming andrnrequires extensive human interaction. We apply multiple clustering techniques, including K-means, hierarchicalrnclustering, and Self-Organizing Map (SOM), to identify similar database objects from heterogeneous databasesrnbased on a combination of features such as object names, documentation, schematic information, data contents,rnand usage patterns. Initial experimental results indicate that our approach performs better than existingrnapproaches in the accuracy of identified interschema relationships. In addition, a prototype system we haverndeveloped provides users a visualization tool for the display of clustering results as well as for the incrementalrnevaluation of candidate solutions.