«IS-THE»-relations in semantic data models: prerequisites and construction rules of «IS-THE»-clusters

«IS-THE»-cluster is a system of classes constructed for restrictions of the same superclass concept; such, that all subclass concepts use the same mapping in the bases of them subtyping. Such mapping can be named discriminating; in fact it is this characteristic of superclass objects that determines their hits in subclasses. At data schema construction people use two different approaches - from top to down and from below upwards. So it is provided both a method of specialization, and a method of generalization for clusters. Ibe first method makes a start from presence of superclass structure and generates a necessary set of subclass structures, the second, having initially structures of subclasses, builds structure for a superclass. When only the structure for a superclass was designed, one can determine need for its specialization and detachment of separate structures for subclasses on the basis of the following signs: - superclass is swelled from optional attributes and relationships; - part of optional attributes and relationships are fixed only if superclass instance satisfies one condition from the fixed set of conditions; - if superclass instance does not satisfy any condition from the fixed set of conditions superclass mandatory attributes and relationships are fixed only. Necessary condition for «IS-THE»-cluster designing from below upwards is presence of classes, whose concepts are close. And further there is a question whether it is necessary to unite these classes in one or several clusters or to leave their untied. «IS-THE»-clusters creation in this case is justified, if: - any business process needs for union of instances of various, but close concepts; - an object of application domain is represented by data in several classes, and it is necessary to restore unity of this object. If at ERM-schema designing one makes a resolve that specialization use is necessary, the designer can adhere to the following technique. 1. To choose a class which will be a superclass of «IS-THE»-cluster, to determine all its mappings to other classes. 2. To group mappings according to their domain: one group is formed from mappings defined on the superclass, other groups of mappings will be defined on some subsets of the superclass. 3. To specify for each subset a kind of detachment (dichotomizing or by basis modification). To determine mapping, which instances form each subset. This mapping is discriminating for created «IS-THE»-cluster. In case of detachment by basis modification should establish a correspondence between each subset and discriminating mapping image. 4. The superclass has still only mappings of its group. Each subset distinguished on the second step is a subclass of constructed «IS-THE»-cluster. This subclass is a domain of all mappings of corresponding group. 5. To define integrity constraints for constructed «IS-THE»-cluster and subclass mappings. If at ERM-schema designing one makes a resolve that generalization use is necessary, the designer can adhere to the following technique. 1. To find the nearest general concept-ancestor to given classes with close concepts. 2. To determine specific difference of each initial concept-descendant and mapping participating in it. If the same mapping appears in all specific differences, and in the case of dichotomizing for one concept-descendant its image is present, and for another concept-descendant is absent, and in case of basis modification concepts-descendants differ in the images, this mapping is discriminating, and one «IS-THE»-cluster is created. 3. If not all variants of discriminating mapping images correspond to initial concepts-descendants, it is necessary to think whether to add to the schema corresponding subclasses. 4. To create a superclass and to transfer on a level of the superclass all common mappings of subclasses, including discriminating mapping. 5. To define integrity constraints for constructed «IS-THE»-cluster and superclass mappings. The techniques submitted in this article are based on ERMM rules of data structurization and integrity constraint definition concerning representation of «IS-THE»-clusters.

Keywords

Authors

References