Database Schema Matching Using Machine Learning with Feature Selection re-publication
by Amihai Motro
Schema matching, the problem of finding mappings between the attributes of two semantically related database schemas, is an important aspect of many database applications such as schema integration, data warehousing, and electronic commerce. Unfortunately, schema matching remains largely a manual, labor-intensive process. Furthermore, the effort required is typically linear in the number of schemas to be matched; the next pair of schemas to match is not any easier than the previous pair. In this paper we describe a system, called Automatch, that uses machine learning techniques to automate schema matching. Based primarily on Bayesian learning, the system acquires probabilistic knowledge from examples that have been provided by domain experts. This knowledge is stored in a knowledge base called the attribute dictionary. When presented with a pair of new schemas that need to be matched (and their corresponding database instances), Automatch uses the attribute dictionary to find an optimal matching. We also report initial results from the Automatch project.