Improving Prototypical Artist Detection by Penalizing Exorbitant Popularity

Abstract

Discovering artists that can be considered as prototypes for particular genres or styles of music is a challenging and interesting task. Based on preliminary work, we elaborate an improved approach to rank artists according to their prototypicality. To calculate such a ranking, we use asymmetric similarity matrices obtained via co-occurrence analysis of artist names on web pages. In order to avoid distortions of the ranking due to ambiguous artist names, e.g. bands whose name equal common speech words (like “Kiss” or “Bush”), we introduce a penalization function. Our approach is demonstrated on a data set containing 224 artists from 14 genres.