Modeling Activity-Driven Music Listening with PACE

TL;DR

The paper addresses how to model listening context by focusing on regular, activity-driven patterns in long-term listening logs. It introduces PACE, which builds multichannel weekly time-series representations and learns a dictionary of interpretable atoms to produce sparse user embeddings. The authors validate PACE by predicting user-reported activities during listening, showing that the embeddings capture meaningful regularities and complement sociological features. This approach offers a scalable, interpretable pathway to contextualize music recommendations around habitual listening behaviors.

Abstract

While the topic of listening context is widely studied in the literature of music recommender systems, the integration of regular user behavior is often omitted. In this paper, we propose PACE (PAttern-based user Consumption Embedding), a framework for building user embeddings that takes advantage of periodic listening behaviors. PACE leverages users' multichannel time-series consumption patterns to build understandable user vectors. We believe the embeddings learned with PACE unveil much about the repetitive nature of user listening dynamics. By applying this framework on long-term user histories, we evaluate the embeddings through a predictive task of activities performed while listening to music. The validation task's interest is two-fold, while it shows the relevance of our approach, it also offers an insightful way of understanding users' musical consumption habits.

Figures (4)

• Figure 1: Aggregation process for a particular user: stream count over the whole dataset (top), weekly scale aggregation (middle), after convolution and normalization (bottom).
• Figure 2: Creating a specific user's embedding.
• Figure 3: Logistic Regression coefficients of the model based purely on PACE embeddings.
• Figure 4: Examples of detected listening patterns: atoms 0, 2 and 27.