Self-organization of conceptual generalities and pattern-directed learning

Abstract

Human expert decision makers can be characterized by their ability to perceive a hypothetical conceptual generality or pattern that is underlying a given collection of objects. The conventional cluster analysis is unable to generate such patterns since its clustering process is far from what the human experts actually do. That is, human experts form some concepts inductively from individual observations based on the conceptual “meaning” which the objects have. In this paper, by introducing an idea of prototype theory from a psychological domain with respect to human concept formation, an algorithm for human classification process is proposed. Based on this, the role of human generalization capability in his classification process is discussed with respect to the background semantic knowledge. The algorithm can be roughly divided into two phases; inductive prototype formation from training examples in a bottom-up fashion, and pattern-directed clustering of the instances being affected by the acquired concepts in a top-down fashion. Using a schematically-modelled example, the algorithm is illustrated with its implemented results. Our modelling method for the human classification process can be utilized for conceptual clustering that classifies a number of unknown objects into a distinguished group being affected by pre-acquired concepts.