AFDGCF: Adaptive Feature De-correlation Graph Collaborative Filtering for Recommendations

TL;DR

This paper investigates feature over-correlation in graph neural network–based collaborative filtering, revealing its prevalence and link to over-smoothing. It proposes the Adaptive Feature De-correlation Graph Collaborative Filtering (AFDGCF), a model-agnostic framework that imposes layer-wise, adaptive penalties on feature correlations to reduce redundancy while preserving collaborative signals. The authors establish a theoretical connection between feature-correlation and smoothing, and validate AFDGCF through extensive experiments across four representative CF models and four public datasets, showing consistent performance gains and reduced representation correlations. The approach also improves training efficiency on some models and scales well to larger datasets, offering a practical path to more robust and personalized GNN-based recommender systems.

Abstract

Collaborative filtering methods based on graph neural networks (GNNs) have witnessed significant success in recommender systems (RS), capitalizing on their ability to capture collaborative signals within intricate user-item relationships via message-passing mechanisms. However, these GNN-based RS inadvertently introduce excess linear correlation between user and item embeddings, contradicting the goal of providing personalized recommendations. While existing research predominantly ascribes this flaw to the over-smoothing problem, this paper underscores the critical, often overlooked role of the over-correlation issue in diminishing the effectiveness of GNN representations and subsequent recommendation performance. Up to now, the over-correlation issue remains unexplored in RS. Meanwhile, how to mitigate the impact of over-correlation while preserving collaborative filtering signals is a significant challenge. To this end, this paper aims to address the aforementioned gap by undertaking a comprehensive study of the over-correlation issue in graph collaborative filtering models. Firstly, we present empirical evidence to demonstrate the widespread prevalence of over-correlation in these models. Subsequently, we dive into a theoretical analysis which establishes a pivotal connection between the over-correlation and over-smoothing issues. Leveraging these insights, we introduce the Adaptive Feature De-correlation Graph Collaborative Filtering (AFDGCF) framework, which dynamically applies correlation penalties to the feature dimensions of the representation matrix, effectively alleviating both over-correlation and over-smoothing issues. The efficacy of the proposed framework is corroborated through extensive experiments conducted with four representative graph collaborative filtering models across four publicly available datasets.

Figures (5)

• Figure 1: Illustration of the GNN-based CF, highlighting the challenges of over-correlation and over-smoothing.
• Figure 2: The figures at the top illustrate the feature correlation (measured by Corr on the left axis) and smoothness (measured by SMV on the right axis) of representations learned by state-of-the-art GNN-based CF models on the Movielens dataset. The bottom figures present the corresponding recommendation performance (measured by NDCG@20).
• Figure 3: An overview of the AFDGCF framework, highlighting its model-agnostic nature and dynamic application of correlation penalties to user and item representations at each layer for enhanced performance.
• Figure 4: Performance comparison w.r.t. different $\alpha$. Using the results of the original model (i.e., $\alpha=0$) as the reference.
• Figure 5: Comparison of GNN-based CF models with different layers before and after applying the AFDGCF framework in terms of NDCG@10, $\textit{Corr}$ and $\textit{SMV}$ on Yelp Dataset.