Mining contour sequences for significant closed patterns

Abstract

Sequential pattern mining in music is a central part of automated music analysis and music generation. This paper evaluates sequential pattern mining on a corpus of Mozarabic chant neume sequences that have been annotated by a musicologist with intra-opus patterns. Significant patterns are discovered in three settings: all closed patterns, maximal closed patterns, and minimal closed patterns. Each setting is evaluated against the annotated patterns using the measures of recall and precision. The results indicate that it is possible to retrieve all known patterns with an acceptable precision using significant closed pattern discovery.