Modelling pattern interestingness in comparative music corpus analysis

Abstract

In computational pattern discovery, pattern evaluation measures select or rank patterns according to their potential interestingness in a given analysis task. Many measures have been proposed to accommodate different pattern types and properties. This paper presents a method and case study employing measures for frequent, characteristic, associative, contrasting, dependent, and significant patterns to model pattern interestingness in a reference analysis, Frances Densmore's study of Teton Sioux songs. Results suggest that interesting changes from older to more recent Sioux songs according to Densmore's analysis are best captured by contrast, dependency, and significance measures.