Dynamic cluster structure and predictive modelling of music creation style distributions

Abstract

We investigate the dynamics of music creation style distributions to understand cultural evolution involving intelligence to create complex artefacts. Previous work suggested that a music creation style can be quantified as statistics describing a generative process of music data, and that the distribution of music creation styles in a society has cluster structure related to the presence of different musical genres. To find patterns in the dynamics of the cluster structure, we analysed statistics of melodies in Japanese popular music data and statistics of audio features in American popular music data. Using statistical modelling methods, we found that intra-cluster dynamics, such as the contraction and the shift of a cluster, as well as inter-cluster dynamics represented by clusters’ relative frequencies, often exhibit notable dynamical modes. Additionally, to compare the individual contributions of these different dynamical aspects for predicting future creation style distributions, we constructed a fitness-based evolutionary model and found that the predictions of cluster frequencies and cluster variances often have comparable contributions. Our results highlight the relevance of intra-cluster dynamics in music style evolution, which have often been overlooked in previous studies. The present methodology can be applied to cultural artefacts whose generative process can be characterized by a discrete probability distribution.

Keywords

• cultural evolution
• statistical learning
• music creation style
• dynamical system
• social dynamics
• trend prediction