Eliciting semantic relations for empirically derived networks

Knowledge elicitation is a critical, yet difficult, process in the development of knowledge-based systems. Pathfinder, a network scaling technique that elicits and represents knowledge in the form of graph structures, has been proposed as a means of overcoming some of the difficulties of other elicitation techniques. However, Pathfinder networks are limited in that their links represent associative, but not semantic, information about conceptual relations. This research addresses the problem of eliciting semantic relations in order to enrich the Pathfinder network representation and increase its potential as a knowledge-elicitation technique. In this paper the SCAN (Sorting, Clustering and Naming) methodology is described and illustrated using links in a network of 20 common concepts. SCAN is also applied to a network of programming concepts. Finally, the methodology is evaluated and compared to related methodologies. Results of these studies indicate that SCAN is a promising link-labelling methodology.