An In-Depth Analysis of Occasional and Recurring Collaborations in Online Music Co-creation

Abstract

The success of online creative communities depends on the will of participants to create and derive content in a collaborative environment. Despite their growing popularity, the factors that lead to remixing existing content in online creative communities are not entirely understood. In this article, we focus on overdubbing , a dyadic collaboration in which one author mixes one new track with an audio recording previously uploaded by another. We study musicians who collaborate regularly, frequently overdubbing each other's songs. Building on frequent pattern–mining techniques, we develop an approach to seek instances of such recurring collaborations in the Songtree community. We identify 43 instances involving two or three members with a similar reputation in the community. Our findings highlight common and different remix factors in occasional and recurring collaborations. Specifically, fresh and less mature songs are generally overdubbed more. Exchanging messages and invitations to collaborate are significant factors only for songs generated through recurring collaborations, whereas author reputation (ranking) and applying metadata tags to songs have a positive effect only in occasional collaborations.

PRESTO: A Recommender of Musical Collaborations based on Heterogeneous Graph Neural Networks

“…Based on different centrality and similarity metrics, they extracted 3 different collaboration clusters representing diverse, regular and absent collaborations. Another study has focused on identifying the main factors that guide the composition of collaborative songs [7].…”

“…Some of the analyses in this line have mainly followed a user-centric view, focusing on solutions for end-users like music recommendation [3,4] or sentiment analysis [5]. Regarding the analysis of musical collaborations, these artist-centric proposals usually follow two prominent lines of research, namely (1) exploratory analyses of the main factors that have guided previous collaborations among musicians [6,7] and (2) approaches for hit-song prediction that consider some collaboration features among the involved artists [8].…”