Addressing popularity discrepancy in collaborative filtering.

Saved in:

Bibliographic Details
Title:	Addressing popularity discrepancy in collaborative filtering.
Authors:	Yu, Cizhou¹ (AUTHOR) czyu22@m.fudan.edu.cn, Li, Dongsheng² (AUTHOR) dongsli@microsoft.com, Gu, Hansu³ (AUTHOR) hansug@acm.org, Zhang, Peng¹ (AUTHOR) zhangpeng_@fudan.edu.cn, Gu, Ning¹ (AUTHOR) ninggu@fudan.edu.cn, Lu, Tun¹ (AUTHOR) lutun@fudan.edu.cn
Source:	Knowledge & Information Systems. Aug2025, Vol. 67 Issue 8, p6525-6551. 27p.
Subjects:	Recommender systems, Popularity
Abstract:	Collaborative filtering (CF) has emerged as the most successful type of recommendation algorithm during the past few decades. However, we observe that CF algorithms often exhibit a popularity discrepancy between user-interacted items and recommended items, e.g., CF algorithms may recommend items that are more popular than the ones the user preferred, especially to those who prefer non-popular items. To address this previously overlooked bias, we make three key contributions: (1) We introduce two novel metrics, PopDis_ED and PopDis_JS, to quantitatively measure popularity discrepancy, providing new perspectives beyond traditional bias indicators; (2) we propose an innovative model-agnostic mutual debiasing (MUDE) framework that uniquely combines a holistic model with a specialized long-tail model through a popularity-aware gating mechanism; (3) comprehensive experiments on four real-world datasets demonstrate that MUDE improves both recommendation accuracy and popularity discrepancy reduction, outperforming state-of-the-art debiasing methods. Moreover, MUDE shows strong generalizability across different types of CF algorithms, making it a practical solution for real-world recommender systems. [ABSTRACT FROM AUTHOR]
	Copyright of Knowledge & Information Systems is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	Collaborative filtering (CF) has emerged as the most successful type of recommendation algorithm during the past few decades. However, we observe that CF algorithms often exhibit a popularity discrepancy between user-interacted items and recommended items, e.g., CF algorithms may recommend items that are more popular than the ones the user preferred, especially to those who prefer non-popular items. To address this previously overlooked bias, we make three key contributions: (1) We introduce two novel metrics, PopDis_ED and PopDis_JS, to quantitatively measure popularity discrepancy, providing new perspectives beyond traditional bias indicators; (2) we propose an innovative model-agnostic mutual debiasing (MUDE) framework that uniquely combines a holistic model with a specialized long-tail model through a popularity-aware gating mechanism; (3) comprehensive experiments on four real-world datasets demonstrate that MUDE improves both recommendation accuracy and popularity discrepancy reduction, outperforming state-of-the-art debiasing methods. Moreover, MUDE shows strong generalizability across different types of CF algorithms, making it a practical solution for real-world recommender systems. [ABSTRACT FROM AUTHOR]
ISSN:	02191377
DOI:	10.1007/s10115-025-02426-1