Development of a structured method for knowledge elicitation
The subject of this thesis is, broadly, knowledge elicitation for knowledge-based, or expert systems. The aims of the research were to investigate the transference of techniques of systems analysis to the knowledge elicitation process, and in so doing, to develop a structured method for knowledge elicitation. The main contributions to the area of knowledge elicitation made by the research are: i) The development of a method which has as a central part of it, the definition of an explicitness boundary, across which and within which all data and processes must be explicit. It is argued that in order to be explicit, the data must be in the form of limited data sets as opposed to continuous data. ii) The development of a method which forces the use of an intermediate representation, thus forcing a logical/physical design split, as in systems analysis for conventional data processing systems. iii) The concern for user independence in the resulting systems. The ability to increase user independence is enhanced by the use of limited data sets, and also by the involvement of designated users of the expert system, and testing of the intermediate representation, during knowledge elicitation. The starting point of the research is the lack of methods for knowledge elicitation, and the pitfalls of existing techniques. Many of the techniques to have emerged from other disciplines such as cognitive psychology are discussed with respect to the concerns of this thesis, and the proposed method. The specific techniques from systems analysis which are applied to knowledge elicitation are data flow analysis, entity-relationship analysis, and entity life cycle modelling. These three techniques form the framework of the method, which starts with a high-level analysis of the domain, and results in an implementation independent representation of the expert domain, equivalent to a logical model in systems analysis and design. The final part of the thesis shows the ease with which the resulting model is translated to two of the most commonly used knowledge representation schemes - production systems and frames.