Mining online music listening trajectories

Abstract

Understanding the listening habits of users is a valuable undertaking for musicology researchers, artists, consumers and online businesses alike. With the rise of Online Music Streaming Services (OMSSs), large amounts of user behavioral data can be exploited for this task. In this paper, we present SWIFT-FLOWS, an approach that models user listening habits in regards to how user attention transitions between artists. SWIFT-FLOWS combines recent advances in trajectory mining, coupled with modulated Markov models as a means to capture both how users switch attention from one artist to another, as well as how users fixate their attention in a single artist over short or large periods of time. We employ SWIFT-FLOWS on OMSSs datasets showing that it provides: (1) semantically meaningful representation of habits; (2) accurately models the attention span of users.