Dual-domain Collaborative Denoising for Social Recommendation

TL;DR

This work tackles noise in social recommendation by introducing DCDSR, a dual-domain denoising framework that operates in both structure and embedding spaces. Its structure-level module denoises the social and interaction graphs through mutual supervision, while its embedding-space module uses dual-domain perturbations and Anchor-InfoNCE to resist cross-domain noise diffusion. Key contributions include a novel preference-guided social denoising mechanism, social-enhanced interaction graph denoising, dual-domain embedding perturbations, and the AC-InfoNCE loss, all trained jointly with the BPR objective. Experiments on six real-world datasets demonstrate robust denoising and consistent performance gains over state-of-the-art baselines, with enhanced resilience to noise attacks. The approach provides practical improvements for robust social recommendation and offers a principled framework for cross-domain denoising in graph-based models.

Abstract

Social recommendation leverages social network to complement user-item interaction data for recommendation task, aiming to mitigate the data sparsity issue in recommender systems. However, existing social recommendation methods encounter the following challenge: both social network and interaction data contain substaintial noise, and the propagation of such noise through Graph Neural Networks (GNNs) not only fails to enhance recommendation performance but may also interfere with the model's normal training. Despite the importance of denoising for social network and interaction data, only a limited number of studies have considered the denoising for social network and all of them overlook that for interaction data, hindering the denoising effect and recommendation performance. Based on this, we propose a novel model called Dual-domain Collaborative Denoising for Social Recommendation ($\textbf{DCDSR}$). DCDSR comprises two primary modules: the structure-level collaborative denoising module and the embedding-space collaborative denoising module. In the structure-level collaborative denoising module, information from interaction domain is first employed to guide social network denoising. Subsequently, the denoised social network is used to supervise the denoising for interaction data. The embedding-space collaborative denoising module devotes to resisting the noise cross-domain diffusion problem through contrastive learning with dual-domain embedding collaborative perturbation. Additionally, a novel contrastive learning strategy, named Anchor-InfoNCE, is introduced to better harness the denoising capability of contrastive learning. Evaluating our model on three real-world datasets verifies that DCDSR has a considerable denoising effect, thus outperforms the state-of-the-art social recommendation methods.

Figures (13)

• Figure 2: Overview of DCDSR.
• Figure 3: Noise diffusion from interaction domain to social domain
• Figure 4: The process of embedding optimizing with InfoNCE and AC-InfoNCE.
• Figure 5: Contributions of each major component in DCDSR.
• Figure 6: Performance comparison of DCDSR with InfoNCE and AC-InfoNCE.