Collaborative Diffusion Model for Recommender System

TL;DR

CDiff4Rec tackles the diffusion-based recommender systems' challenge of preserving personalization while leveraging rich item content. It converts item review features into pseudo-users and identifies two sets of personalized neighbors (real and pseudo) to guide diffusion via an attention-based collaborative signals aggregator, yielding refined user representations. Empirical results on three public datasets show CDiff4Rec consistently outperforms baselines, with a favorable accuracy–efficiency balance, and ablations validate the contribution of both pseudo-users and neighbor signals. This approach broadens diffusion models for recommendation by integrating content-derived signals and neighbor-aware reconstruction, potentially enabling more nuanced and scalable personalization in sparse settings.

Abstract

Diffusion-based recommender systems (DR) have gained increasing attention for their advanced generative and denoising capabilities. However, existing DR face two central limitations: (i) a trade-off between enhancing generative capacity via noise injection and retaining the loss of personalized information. (ii) the underutilization of rich item-side information. To address these challenges, we present a Collaborative Diffusion model for Recommender System (CDiff4Rec). Specifically, CDiff4Rec generates pseudo-users from item features and leverages collaborative signals from both real and pseudo personalized neighbors identified through behavioral similarity, thereby effectively reconstructing nuanced user preferences. Experimental results on three public datasets show that CDiff4Rec outperforms competitors by effectively mitigating the loss of personalized information through the integration of item content and collaborative signals.

Figures (1)

• Figure 1: The overview of the proposed Collaborative Diffusion Model for Recommender System (CDiff4Rec).