Concept Membership Modeling Using a Choquet Integral

Abstract

Imprecise and subjective concepts, as e.g. promising students, may be used within data mining tasks or database queries to faithfully describe data properties of interest. However, defining these concepts is a demanding task for the end-user. We thus provide a strategy, called CHOCOLATE, that only requires the user to give a tiny subset of data points that are representative of the concept he/she has in mind, and that infers a membership function from them. This function may then be used to retrieve, from the whole dataset, a ranked list of points that satisfy the concept of interest. CHOCOLATE relies on a Choquet integral to aggregate the relevance of individual attribute values among all the representative points as well as the representativity of sets of such attribute values. As a consequence, a valuable property of the proposed approach is that it is able to both capture properties shared by most of the user-selected representative data points as well as specific properties possessed by only one specific representative data point.