|dc.description.abstract||Any expert system shell that performs with the generic task of hierarchical classificiation must deal explicitly with the issues of knowledge representations, control strategies, inductive learning, and ways of handling uncertainty, ambiguity, and contradictions. This resesarch is mainly concerned about the creation of the expert system shell HICLASS. Aspects crucial to this task are challenged from btoh a theoretical and an implementational point of view.
The principles of generic tasks and hierarchical classification are described. Important concepts of HICLASS are introducted, followed by a detailed description of the knowledge representation and local control strategies developed for the system, including a discussion of special problems and respective solutions. IT is described how HICLASS handles uncertainty. Important issues like concluding values, explanation, learning, incorporating metaknowledge, and the global control strategy of HICLASS are discussed. Then, the actual implementation of the table editor HIEDIT as well as HICLASS is described in detail. It is show that HICLASS is a genuine tool for the generic task for hierarchical classification. The system is compared to two well-known tools for hierarchical classification. Using the ideas raised for HICLASS, the development of a hierarchical hypothesis matcher, HIHYPO, is proposed. Essential features of HIHYPO are introducted. A theoretic overview about algorithms for inductive learning is followed by the description of an inductive learning algorithm developed for HIHYPO. Appendix B provides an overview about software engineering methods, and a discussioin about methods actually used to create the HICLASS package. In Appendix C, the definitions of all modules developed for the package are shown.||en_US