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

2026-05-11

AI Summary

This work addresses the tendency of graph neural networks (GNNs) in collaborative filtering to amplify popularity bias, which often leads to the neglect of long-tail items. To mitigate this issue, the authors propose the DPAA framework, which incorporates embedding-aware dynamic interaction weighting and inter-layer weight adjustment during message passing. This approach explicitly integrates representation-level popularity signals into the aggregation process while employing a smooth transition strategy and high-order neighborhood enhancement to effectively suppress bias propagation. Experimental results demonstrate that DPAA significantly outperforms existing debiasing methods on both real-world and semi-synthetic datasets, achieving substantial improvements in long-tail item coverage and recommendation fairness.

Abstract

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.

Problem

• popularity bias
• graph neural networks
• collaborative filtering
• message passing
• long-tail items

Innovation

• popularity bias
• graph neural networks
• message passing
• collaborative filtering
• debiasing