Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Modeling Musical Genre Trajectories through Pathlet Learning

Lilian Marey, Charlotte Laclau, Bruno Sguerra, Tiphaine Viard, Manuel Moussallam

TL;DR

This work introduces pathlet learning, a dictionary-learning-based framework to model the evolution of users' musical tastes by extracting recurring small-genres trajectories (pathlets) from listening histories. By constructing trajectory embeddings and learning a pathlet dictionary on a genre-graph, the approach provides interpretable representations that capture appearance and disappearance of genres over time, and improves predictive metrics like average total variation and new-classes emergence. The method is validated on a Deezer dataset and a Last.fm dataset, showing quantitative gains over baselines and revealing qualitative insights into genre dynamics, co-listening patterns, and diversity considerations for recommender systems. The released code and Deezer-derived dataset facilitate reproducibility, further research into explainable trajectory modeling, and potential applications in long-term taste modeling, playlist curation, and diversity-aware recommendations.

Abstract

The increasing availability of user data on music streaming platforms opens up new possibilities for analyzing music consumption. However, understanding the evolution of user preferences remains a complex challenge, particularly as their musical tastes change over time. This paper uses the dictionary learning paradigm to model user trajectories across different musical genres. We define a new framework that captures recurring patterns in genre trajectories, called pathlets, enabling the creation of comprehensible trajectory embeddings. We show that pathlet learning reveals relevant listening patterns that can be analyzed both qualitatively and quantitatively. This work improves our understanding of users' interactions with music and opens up avenues of research into user behavior and fostering diversity in recommender systems. A dataset of 2000 user histories tagged by genre over 17 months, supplied by Deezer (a leading music streaming company), is also released with the code.

Modeling Musical Genre Trajectories through Pathlet Learning

TL;DR

This work introduces pathlet learning, a dictionary-learning-based framework to model the evolution of users' musical tastes by extracting recurring small-genres trajectories (pathlets) from listening histories. By constructing trajectory embeddings and learning a pathlet dictionary on a genre-graph, the approach provides interpretable representations that capture appearance and disappearance of genres over time, and improves predictive metrics like average total variation and new-classes emergence. The method is validated on a Deezer dataset and a Last.fm dataset, showing quantitative gains over baselines and revealing qualitative insights into genre dynamics, co-listening patterns, and diversity considerations for recommender systems. The released code and Deezer-derived dataset facilitate reproducibility, further research into explainable trajectory modeling, and potential applications in long-term taste modeling, playlist curation, and diversity-aware recommendations.

Abstract

The increasing availability of user data on music streaming platforms opens up new possibilities for analyzing music consumption. However, understanding the evolution of user preferences remains a complex challenge, particularly as their musical tastes change over time. This paper uses the dictionary learning paradigm to model user trajectories across different musical genres. We define a new framework that captures recurring patterns in genre trajectories, called pathlets, enabling the creation of comprehensible trajectory embeddings. We show that pathlet learning reveals relevant listening patterns that can be analyzed both qualitatively and quantitatively. This work improves our understanding of users' interactions with music and opens up avenues of research into user behavior and fostering diversity in recommender systems. A dataset of 2000 user histories tagged by genre over 17 months, supplied by Deezer (a leading music streaming company), is also released with the code.
Paper Structure (26 sections, 9 equations, 6 figures, 1 table)

This paper contains 26 sections, 9 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related works
  3. Background and Motivation
  4. Dictionary and Pathlet Learning
  5. From a Geographical Context to User Trajectories
  6. Pathlet Learning for User Trajectories
  7. Problem Framing and Notations
  8. Defining User Trajectories
  9. Building Trajectories
  10. Candidate Sets
  11. Learning the Dictionary
  12. Defining Graph and Paths' Representations
  13. Computing the Dictionary
  14. Building Trajectory Embeddings
  15. Evolution of Musical Taste: Experimental Protocol
  16. ...and 11 more sections

Figures (6)

  • Figure 1: Our methodology for users preferences modeling. Music genre trajectories are extracted from user listening histories and processed using the pathlet learning algorithm. This highlights recurring patterns in user behavior, which are then used to generate explicit trajectory embeddings that predict future genre allocation.
  • Figure 2: Users' genre listening variation types between 2 successive periods with respect to intra-variability. While appearance and disappearance correspond to emergence or abandonment of genres, persistence rely on variations of listening rate for a genre listened during 2 successive periods.
  • Figure 3: Reconstruction error of $X$ over the number of sampled trajectories.
  • Figure 4: Sensitive Analysis: Impact of $\lambda$ on dictionary performances.
  • Figure 5: Pathlet interpretations : interactions between genres (top) and pathlet diversity (bottom)
  • ...and 1 more figures