Breaking the Loop: An Empirical Comparison of Strategies for Novelty and Freshness in YouTube Music
Abstract
Continuously trained ranking models in music recommenders fall into feedback loops where previously consumed items dominate recommendations. This suppresses two distinct content classes: new releases (temporal freshness) and unlistened catalog items (novelty). Industry practitioners have a wide menu of interventions available, ranging from serving-time heuristics, training-data reweighting, architectural debiasing, to uncertainty-driven exploration, Each of which are well understood in academic settings. But live systems offer challenges with continuously ingested content, interconnected components, and practical limitations that counteract the findings from academic research.
We report online A/B results for six interventions and one combination experiment across four conceptual layers (serving, training, architecture, exploration) on the YouTube Music homepage. All interventions modify the ranking model or the serving layer that consumes it scores; candidate generation and other upstream components are held fixed. Architectural debiasing produced the largest reduction in popularity dominance but penalized new releases. Uncertainty-driven exploration with an SNGP head drove +7.7\% Day-1 new-release engagement at no measurable homepage cost, at the expense of artist diversity. Stacking the bias tower with uncertainty-weighted loss composed additively on popularity redistribution and artist diversity, but absorbed the singleton new-release lift. We close with practitioner-facing takeaways on which layer to intervene at, and the hidden costs of each choice.
We report online A/B results for six interventions and one combination experiment across four conceptual layers (serving, training, architecture, exploration) on the YouTube Music homepage. All interventions modify the ranking model or the serving layer that consumes it scores; candidate generation and other upstream components are held fixed. Architectural debiasing produced the largest reduction in popularity dominance but penalized new releases. Uncertainty-driven exploration with an SNGP head drove +7.7\% Day-1 new-release engagement at no measurable homepage cost, at the expense of artist diversity. Stacking the bias tower with uncertainty-weighted loss composed additively on popularity redistribution and artist diversity, but absorbed the singleton new-release lift. We close with practitioner-facing takeaways on which layer to intervene at, and the hidden costs of each choice.
Research Areas
