Saved in:
Bibliographic Details
Main Authors: Chan, Audrey, Labbé, Aaron, Lavoie, Jacob, Bannister, Jordan, Tchango, Arsène Fansi, Lajoie, Guillaume, Charlin, Laurent
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.28810
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918528160890880
author Chan, Audrey
Labbé, Aaron
Lavoie, Jacob
Bannister, Jordan
Tchango, Arsène Fansi
Lajoie, Guillaume
Charlin, Laurent
author_facet Chan, Audrey
Labbé, Aaron
Lavoie, Jacob
Bannister, Jordan
Tchango, Arsène Fansi
Lajoie, Guillaume
Charlin, Laurent
contents Functional music applications, from consumer focus and sleep aids to clinical interventions, share a distinctive recommendation problem: success is defined by the listener's affective state, but online experimentation on emotion is ethically constrained, particularly for clinical populations who cannot reliably skip a song or report distress. We describe AMRS, the Affective Music Recommendation System deployed on LUCID's health-and-wellness platforms, which serve clinical users (primarily older adults with neurocognitive conditions) and consumer-wellness users across energize, focus, calm, and sleep modes. AMRS is built around a rollout-based world model: a causal transformer trained on logged listening data to jointly predict engagement, binary rating, and self-reported valence and arousal. The world model serves both as an in-silico simulator for offline policy training and as a stress-testing tool before deployment. A recommender policy initialized by behaviour cloning is fine-tuned offline with Direct Preference Optimization (DPO) against a configurable multi-objective utility function. Under a strict cold-start protocol, the world model predicts both behavioural and affective signals with usable fidelity; DPO improves predicted valence and arousal over the cloned baseline while maintaining a similar diversity profile and avoiding the distributional collapse produced by greedy optimization. We position the work as an early deployed validation of a methodology for affective recommendation when online experimentation is ethically untenable.
format Preprint
id arxiv_https___arxiv_org_abs_2605_28810
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Affective Music Recommendation: A Rollout-Based World Model for Offline Preference Optimization
Chan, Audrey
Labbé, Aaron
Lavoie, Jacob
Bannister, Jordan
Tchango, Arsène Fansi
Lajoie, Guillaume
Charlin, Laurent
Machine Learning
Information Retrieval
Sound
Functional music applications, from consumer focus and sleep aids to clinical interventions, share a distinctive recommendation problem: success is defined by the listener's affective state, but online experimentation on emotion is ethically constrained, particularly for clinical populations who cannot reliably skip a song or report distress. We describe AMRS, the Affective Music Recommendation System deployed on LUCID's health-and-wellness platforms, which serve clinical users (primarily older adults with neurocognitive conditions) and consumer-wellness users across energize, focus, calm, and sleep modes. AMRS is built around a rollout-based world model: a causal transformer trained on logged listening data to jointly predict engagement, binary rating, and self-reported valence and arousal. The world model serves both as an in-silico simulator for offline policy training and as a stress-testing tool before deployment. A recommender policy initialized by behaviour cloning is fine-tuned offline with Direct Preference Optimization (DPO) against a configurable multi-objective utility function. Under a strict cold-start protocol, the world model predicts both behavioural and affective signals with usable fidelity; DPO improves predicted valence and arousal over the cloned baseline while maintaining a similar diversity profile and avoiding the distributional collapse produced by greedy optimization. We position the work as an early deployed validation of a methodology for affective recommendation when online experimentation is ethically untenable.
title Affective Music Recommendation: A Rollout-Based World Model for Offline Preference Optimization
topic Machine Learning
Information Retrieval
Sound
url https://arxiv.org/abs/2605.28810