How to describe the spatial near-far relations among concepts?

Abstract

Research in cognitive science shows that when humans learn concepts, relation-based categories are easier to learn than feature-based categories, and humans can grasp the essential difference of things in different relations and distinguish a certain degree of diversity. For example, in terms of color, blue sea is closer to blue sky than red coral. How can the machine understand the near-far relations among different concepts? Drawing on the idea of granular computing and starting from the perspective of relation, this study proposes a method to describe the spatial near-far relations among different concepts. First, the eigenvalue representation methods of the matrix are given, and on this basis, the salient feature space of the symmetric matrix is proposed. Then, we propose a salient representation method for category, and through qualitative analysis from the perspective of consistency and near-far relations, we find that the salient representation can represent categories from the perspective of attribute and object respectively. Finally, we verify the validity of salient relation representation through second-order isomorphism between intent and extent of concept quantitatively.