Aligning Recommendations with User Popularity Preferences

TL;DR

This paper introduces a new framework and method to measure and mitigate popularity bias in recommender systems, improving alignment with individual user preferences without sacrificing recommendation quality.

Contribution

It proposes Popularity Quantile Calibration for measuring bias and SPREE, an activation steering method, to enhance user-specific popularity alignment in recommendations.

Findings

SPREE improves user-level popularity alignment across datasets.

SPREE maintains recommendation quality while reducing popularity bias.

The framework effectively quantifies misalignment between user preferences and recommendations.

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.