Debiasing Message Passing to Mitigate Popularity Bias in GNN-based Collaborative Filtering

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Islam, Md Aminul, Faruk, Ahmed Sayeed, Medya, Sourav, Zheleva, Elena
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911672241750016
author Islam, Md Aminul
Faruk, Ahmed Sayeed
Medya, Sourav
Zheleva, Elena
author_facet Islam, Md Aminul
Faruk, Ahmed Sayeed
Medya, Sourav
Zheleva, Elena
contents Collaborative filtering (CF) models based on graph neural networks (GNNs) achieve strong performance in recommender systems by propagating user-item signals over interaction graphs. However, they are highly susceptible to popularity bias, since skewed interaction distributions and repeated message passing across high-order neighborhoods amplify the influence of popular items while suppressing long-tail ones. Existing debiasing approaches, including re-weighting objectives, regularization, causal methods, and post-processing, are less effective in GNN-based settings because they do not directly counteract bias propagated through the aggregation process, and recent in-aggregation weighting methods often rely on static heuristics or unstable embedding estimates. We propose Debiasing Popularity Amplification in Aggregation (DPAA), a popularity debiasing framework for GNN-based CF that integrates adaptive, embedding-aware interaction weighting and layer-wise weighting directly into message passing. DPAA assigns interaction-level weights from a representation-aware popularity signal, stabilized by a smooth transition from pre-trained to evolving model embeddings during training. It further introduces a layer-wise weighting that amplifies higher-order neighborhoods, surfacing long-range interactions with diverse and underexposed items. Experiments on real-world and semi-synthetic datasets show that DPAA outperforms state-of-the-art popularity-bias correction methods for GNN-based CF.
format Preprint
id arxiv_https___arxiv_org_abs_2605_11145
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Debiasing Message Passing to Mitigate Popularity Bias in GNN-based Collaborative Filtering
Islam, Md Aminul
Faruk, Ahmed Sayeed
Medya, Sourav
Zheleva, Elena
Information Retrieval
Machine Learning
Collaborative filtering (CF) models based on graph neural networks (GNNs) achieve strong performance in recommender systems by propagating user-item signals over interaction graphs. However, they are highly susceptible to popularity bias, since skewed interaction distributions and repeated message passing across high-order neighborhoods amplify the influence of popular items while suppressing long-tail ones. Existing debiasing approaches, including re-weighting objectives, regularization, causal methods, and post-processing, are less effective in GNN-based settings because they do not directly counteract bias propagated through the aggregation process, and recent in-aggregation weighting methods often rely on static heuristics or unstable embedding estimates. We propose Debiasing Popularity Amplification in Aggregation (DPAA), a popularity debiasing framework for GNN-based CF that integrates adaptive, embedding-aware interaction weighting and layer-wise weighting directly into message passing. DPAA assigns interaction-level weights from a representation-aware popularity signal, stabilized by a smooth transition from pre-trained to evolving model embeddings during training. It further introduces a layer-wise weighting that amplifies higher-order neighborhoods, surfacing long-range interactions with diverse and underexposed items. Experiments on real-world and semi-synthetic datasets show that DPAA outperforms state-of-the-art popularity-bias correction methods for GNN-based CF.
title Debiasing Message Passing to Mitigate Popularity Bias in GNN-based Collaborative Filtering
topic Information Retrieval
Machine Learning
url https://arxiv.org/abs/2605.11145