This paper defines a method for decomposing a large data model into a hierarchy of models of manageable size. The purpose of this is to (a) improve user understanding and (b) simplify documentation and maintenance. Firstly, a set of principles is defined which prescribe the characteristics of a "good" decomposition. These principles may be used to evaluate the quality of a decomposition and to choose betweenalternatives. Based on these principles, a manual procedure is described which can be used by a human expert to produce a relatively optimal clustering. Finally, a genetic algorithm is described which automatically finds an optimal decomposition. A key differentating factor between this and previous approaches is that it is soundly basedo n principles of human information processing-this ensures that data models are clustered in a way that can be most efficiently processed by the human mind.
展开▼