Abstract
When a dataset involves multiple classes, there is often a need to
express the key contrasting features among these classes in humanly
understandable terms, that is, to profile the classes. Commonly, one
class is contrasted from the rest by aggregating the latter into a
pseudo-class; alternatively, classes are treated separately without
coordinating their profiles with those of the other classes. We
introduce the concise all pairs profiling (CAPP) method for concise,
intelligible, and approximate profiling of large classifications. The
method compares all classes pairwise and then minimizes the overall
number of features needed to guarantee that each pair of classes is
contrasted by at least one feature. Then each class profile gets its
own minimized list of features, annotated with how these features
contrast the class from the others. Significant applications to
social and natural science are demonstrated.
full paper