In this paper we present a research dealing with musical improvisation practices on TikTok as part of a project devoted to the development of a music improvisation software. Finding musicians to experiment with led us to explore the musical collaborations encouraged by the platform. The duo function appears to be one of its most important feature from this point of view. Tracking relationships between musicians who duet with each other leads to the definition of musical communities that form on the network. These online communities pose new problems to the classic ethnographic approach to communities, and their study can benefit from new methods involving web scraping techniques. This article shows how to use these techniques in the context of TikTok and analyses some properties of the graphs that can be drawn from web scraping data collected within the musical communities that are formed thanks to the duet function. Certain properties of the graph, such as source and sink nodes, are highlighted and linked to musical roles such as soloist and accompanist.