Aligning Recommendations with User Popularity Preferences

2026-04-01

AI Summary

This work addresses the prevalence of popularity bias in recommender systems, which often leads to over-recommendation of popular items while overlooking users’ individual preferences for either niche or mainstream content. To mitigate this issue, the authors propose SPREE, a novel inference-time debiasing method that dynamically adjusts user representations through activation manipulation to achieve personalized debiasing. Departing from conventional approaches, SPREE introduces a popularity quantile calibration framework grounded in the alignment between users and the recommender, enabling precise quantification of preference misalignment. It further identifies popularity directions in the representation space and adaptively modulates both the strength and orientation of alignment. Extensive experiments demonstrate that SPREE significantly enhances user-level popularity alignment across multiple datasets without compromising overall recommendation performance.

Abstract

Popularity bias is a pervasive problem in recommender systems, where recommendations disproportionately favor popular items. This not only results in "rich-get-richer" dynamics and a homogenization of visible content, but can also lead to misalignment of recommendations with individual users' preferences for popular or niche content. This work studies popularity bias through the lens of user-recommender alignment. To this end, we introduce Popularity Quantile Calibration, a measurement framework that quantifies misalignment between a user's historical popularity preference and the popularity of their recommendations. Building on this notion of popularity alignment, we propose SPREE, an inference-time mitigation method for sequential recommenders based on activation steering. SPREE identifies a popularity direction in representation space and adaptively steers model activations based on an estimate of each user's personal popularity bias, allowing both the direction and magnitude of steering to vary across users. Unlike global debiasing approaches, SPREE explicitly targets alignment rather than uniformly reducing popularity. Experiments across multiple datasets show that SPREE consistently improves user-level popularity alignment while preserving recommendation quality.

Problem

• popularity bias
• recommender systems
• user preference alignment
• recommendation homogenization
• rich-get-richer dynamics

Innovation

• popularity bias
• user alignment
• activation steering
• sequential recommendation
• inference-time mitigation