Graph-based Diffusion Model for Collaborative Filtering

TL;DR

GDMCF extends diffusion-based recommender systems to the user–item bipartite graph to capture higher-order collaborative signals. It tackles noise heterogeneity with a multi-level corruption scheme (discrete structure and continuous features) and mitigates relation explosion via a user-active guided generation strategy during inference, supported by a graph-based denoising network. The method combines a unified graph construction and a GCN-based denoiser, optimizing with a diffusion loss plus alignment loss to recover a clean graph for ranking. Empirical results on ML-1M, Yelp, and Amazon-Book show state-of-the-art performance and improved efficiency, with ablations validating the contribution of each component and demonstrating robustness on long-tail data.

Abstract

Recently, diffusion-based recommendation methods have achieved impressive results. However, existing approaches predominantly treat each user's historical interactions as independent training samples, overlooking the potential of higher-order collaborative signals between users and items. Such signals, which encapsulate richer and more nuanced relationships, can be naturally captured using graph-based data structures. To address this limitation, we extend diffusion-based recommendation methods to the graph domain by directly modeling user-item bipartite graphs with diffusion models. This enables better modeling of the higher-order connectivity inherent in complex interaction dynamics. However, this extension introduces two primary challenges: (1) Noise Heterogeneity, where interactions are influenced by various forms of continuous and discrete noise, and (2) Relation Explosion, referring to the high computational costs of processing large-scale graphs. To tackle these challenges, we propose a Graph-based Diffusion Model for Collaborative Filtering (GDMCF). To address noise heterogeneity, we introduce a multi-level noise corruption mechanism that integrates both continuous and discrete noise, effectively simulating real-world interaction complexities. To mitigate relation explosion, we design a user-active guided diffusion process that selectively focuses on the most meaningful edges and active users, reducing inference costs while preserving the graph's topological integrity. Extensive experiments on three benchmark datasets demonstrate that GDMCF consistently outperforms state-of-the-art methods, highlighting its effectiveness in capturing higher-order collaborative signals and improving recommendation performance.

Figures (7)

• Figure 1: Recommender systems involve various forms of heterogeneous noise (continuous noise and discrete noise). As the number of users and items increases, the complexity of their interactions can grow explosively.
• Figure 2: The Framework of GDMCF.
• Figure 3: User-Active Guided Diffusion Modeling. In the inference process, GDMCF iteratively adds edges from $\mathcal{G}^{T^{\prime}}$ based on the original graph's degree distribution. The probabilities (0.3, 0.4, 0.2) determine if a user node is activated at step $t$, with only activated nodes' edges retained (e.g., node 2 has a 0.4 probability of retaining its edges).
• Figure 4: Influence of the number of GCN layers $l$.
• Figure 5: Influence of the loss hyperparameter $\lambda_1$.