Session-Based Segmentation of Music Listening Behaviour on Digital Platforms Using K-Means Clustering

Abstract

Gaining insight into digital music listening behaviour is essential for optimizing personalization and enhancing user engagement. Instead of using traditional aggregated data, we employed a session-based methodology to reveal detailed interaction dynamics. A "session" was simply defined as a set of tracks played sequentially within a 30-minute window. By processing playback data into features and applying K-Means clustering, we successfully delineated four unique listener profiles. These profiles differ significantly across three dimensions: session length, skipping frequency, and listening time patterns. The practical relevance of these findings is high, directly informing platform recommender systems, feature strategy, and marketing efforts, while also contributing to the academic case for session-based user modelling.