Development of a methodology for optimizing the elicited knowledge
Abstract
Knowledge elicitation is the first step in building expert systems, and it is a major bottleneck in the construction of expert systems. In this research, a conceptual framework and methodology is presented for selecting knowledge elicitation methods. This methodology provides an integrated framework of knowledge structures, knowledge elicitation methods, cognitive factors of human abilities, and task types. A statistical nested factorial design is utilized with three tasks (diagnosis, debugging and interpretation), four knowledge elicitation methods (protocol, interview, induction and repertory grid) and ten cognitive factors. Twenty-four subjects were used in the experiment, and five hypotheses were tested. The Multivariate Analysis of Variance (MANOVA) and Student-Newman-Keuls (SNK) test indicate that there is no significant difference among the four knowledge elicitation methods in acquiring the expert's knowledge in diagnosis or interpretation tasks. However for the debugging task, the interview and induction methods yield superior knowledge than do the protocol or repertory grid methods. The induction and repertory grid method take less time to obtain expert knowledge than do the interview or protocol methods. The interview method is best when acquiring important data from an expert in debugging or interpretation tasks. The repertory grid and induction are the most efficient methods for acquiring the knowledge needed for debugging and interpretation tasks. The expert's cognitive abilities significantly affect the percentage of knowledge acquired from experts. Based on these findings, a Matching Index for combining tasks, knowledge elicitation methods and cognitive abilities is derived. This matching index maximizes the elicited knowledge by selecting the most appropriate method of knowledge elicitation for specific tasks and also selecting the best individuals for this knowledge elicitation.
Degree
Ph.D.
Advisors
Salvendy, Purdue University.
Subject Area
Industrial engineering|Artificial intelligence|Systems design
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.