Quantify music artist similarity based on style and mood

Abstract

Music artist similarity has been an active research topic in music information retrieval for a long time since it is especially useful for music recommendation and organization. However, it is a difficult problem. The similarity varies significantly due to different artistic aspects considered and most importantly, it is hard to quantify. In this paper, we propose a new framework for quantifying artist similarity. In the framework, we focus on style and mood aspects of artists whose descriptions are extracted from the authoritative information available at the All Music Guide website. We then generate style--mood joint taxonomies using hierarchical co-clustering algorithm, and quantify the semantic similarities between the style/mood terms based on the taxonomy structure and the positions of these terms in the taxonomies. Finally we calculate the artist similarities according to all the style/mood terms used to describe them. Experiments are conducted to show the effectiveness of our framework.