This paper describes a computer based approach to comparing data obtained for Knowledge Based Systems via established but very varied knowledge elicitation (KE) techniques. It describes not only the detailed comparison of diflerent KE methods (in this case 'scaling' and 'non-scaling ') but also investigates the use of 'demonstration ' or 'evaluation' systems, as a variation on the more established rapid prototyping approaches to the elicitation and evaluation of knowledge for KBS construction, in this case by focusing upon the quality and relevance of the elicited knowledge from the perspective of the expert himsev Preliminary results from the study reported here suggest that non-scaling methods produce a greater amount of raw data than scaling methods, and that this data is less likely to require correction or modiJication for inclusion within a Knowledge Based System. However, the results also indicate that non-scaling derived data is more likely than scaling derived data data to be rejected outright.

Although knowledge elicitation, the process of extracting knotvledgefrom human experts to be incorporated into a knowledge-basedsystem, has been the subject of some notable studies, less attention has been paid to the methods of analysing the raw data once it has been extractedfrom the expert. When knowledge elicitation sessions are interview-based, the resultant form of raw data is usually a transcript of the interviewee's utterances. This paper describes an investigation into the preliminary stage of analysing such transcripts. It outlines the development of an approach to eliminate unnecessary detailfrom interview transcripts, thus enabling attention to be focused upon the remaining, more relevant data via a simple technique based upon cheap and readily available technology. The paper then outlines a rapid-prototyping approach for evaluating this method, the results of which werefelt to be very encouraging.

No abstract

No abstract