The problem of unique name violations in database integration

Abstract

When multiple database schemas are integrated, there are often conflicts in the naming of attributes within the schemas. These conflicts must be detected and resolved prior to successful integration of the schemas. This thesis describes a method for automatically detecting such naming conflicts, which adapts and enhances a method for detecting similar conflicts in (mathematical) model integration. The method relies on the representation of semantic information, not found in data dictionaries, about the data elements or attributes present in the various schemas. This information about data elements is then used by mechanical inference procedures to automatically determine whether two distinctly named elements in fact represent the same object (the synonym problem), or if data elements with the same name in different schemas actually represent different objects (the homonym problem). The expected accuracy and errors of these procedures, and results obtained from a set of experiments on the use of this method, are also presented.