Aligning Recommendations with User Popularity Preferences
Summary: arXiv:2604.01036v1 Announce Type: cross
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.
Understanding Popularity Bias
Popularity bias occurs when recommender systems tend to favor items that are already popular, leading to a feedback loop that enhances the visibility of these items while marginalizing niche content. This can create issues for users who have unique preferences that do not align with mainstream popularity.
Introducing Popularity Quantile Calibration
To address the challenges posed by popularity bias, our study introduces a new framework known as Popularity Quantile Calibration. This framework serves to measure the extent of misalignment between a user’s historical preferences for popular or niche items and the popularity of the recommendations they receive. By quantifying this misalignment, we can gain insights into how well the recommender system aligns with individual user preferences.
SPREE: A Solution for Popularity Alignment
SPREE, which stands for Steering Popularity Recommendations for Enhanced Experience, is designed to mitigate the effects of popularity bias in a more personalized manner. Unlike traditional global debiasing approaches that uniformly reduce the visibility of popular items, SPREE focuses on enhancing the alignment between user preferences and recommendations.
Key features of SPREE include:
- Adaptive Steering: SPREE adapts the steering of model activations based on each user’s individual popularity bias, allowing for a more personalized recommendation experience.
- Direction and Magnitude Variation: The method allows both the direction and magnitude of the steering to vary across different users, ensuring that recommendations are tailored to their unique preferences.
- Focus on Alignment: Instead of merely reducing the popularity of recommendations, SPREE explicitly targets alignment with user preferences, aiming to improve user satisfaction while maintaining recommendation quality.
Experimental Validation
Through experiments conducted across multiple datasets, SPREE has consistently shown improvements in user-level popularity alignment. The results indicate that users are more likely to receive recommendations that resonate with their individual preferences, whether for popular or niche items. Furthermore, these improvements have been achieved without compromising the overall quality of the recommendations provided.
Conclusion
In conclusion, the introduction of Popularity Quantile Calibration and the SPREE method represents a significant advancement in addressing the issue of popularity bias in recommender systems. By focusing on user-recommender alignment, we can create a more equitable and satisfying recommendation experience for all users, ensuring that both popular and niche content receive the visibility they deserve.