Acquiring descriptive knowledge for classification and identification

Abstract

During the past decade, numerous real world knowledge-based systems have been built for the purpose of identification. Although most identification systems are based on the ability to observe and describe, few systems adress one of the first steps in knowledge acquisition which is how to acquire descriptions. Collecting this descriptive knowledge (observed facts) requires that a descriptive model (observable facts) has been previously defined. In addition, experience shows that the model depends on the goal which is pursued. In this paper, we present a tool and a methodology for the acquisition of the descriptive knowledge and the corresponding model which was designed primarily for identification. To achieve this goal, we have first used induction and have ran into redhibitory problems due to some limitations of this technology for processing incomplete descriptions. We present how we have stretched the technology in a case-based reasoning fashion to overcome these limitations. The tools and methodology have been developped and validated in the context of several real world applications.