Conceptual Clustering of Complex Objects: A Generalization Space based Approach

A key issue in learning from observations is to build a classification of given objects or situations. Conceptual clustering methods address this problem of recognizing regularities among as set of objects that have not been pre-classified, so as to organize them into a hierarchy of concepts. Early approaches have been limited to unstructured domains, in which objects are described by fixed sets of attribute-value pairs. Recent approaches in structured domains use a first order logic based representation to represent complex objects. The problem addressed in this paper is to provide a basis for the analysis of complex objects clustering represented using conceptual graphs formalism. We propose a new clustering method that extracts a hierarchical categorization of the provided objects from an explicit space of concepts hierarchies, called Generalization Space. We give a general algorithm and expose several complexity factors.