The EGG/YOLK Reliability Hierarchy: Semantic Data Integration Using Sorts With Prototypes

Abstract

Integration of disparate heterogeneous databases requires translation of types. Because a type in one system often has no exact counterpart in the others, fully reliable integration requires deep understanding of the subject domain, with conceptual analysis of type meanings. So far, reliable translation has had to be done by hand. In practice, few types are so crucial as to require full reliability. The <b>EGG/YOLK</b> hierarchy ranks types by the tolerable rashness in translation, based on prototypes in each type. Each defined class (<b>EGG</b>) has a subclass of typical members (<b>YOLK</b>) defined. We exploit Cui, Cohn and Randell's Qualitative Spatial Simulation program to create the hierarchy of all possible relations between source and target <b>EGG/YOLK</b> types, ranked by reliability. Our eventual ranking is based on a poset combining four different preference criteria.