Invariance Matters: Empowering Social Recommendation via Graph Invariant Learning

TL;DR

This paper tackles the challenge of noise in social graphs used for recommender systems by reframing social denoising as an invariant learning problem. The authors introduce SGIL, which first simulates multiple noisy social environments and then learns user preferences that remain stable across these environments by minimizing an invariant risk with adversarially generated diversity. The method combines a LightGCN-S-based encoder, environment-specific empirical risk, and a variance-based invariance regularizer, yielding superior performance over state-of-the-art baselines across three benchmarks and showing robustness to varying noise and sparsity. SGIL’s invariance-centric approach offers a principled path to robust social recommendation in the presence of noisy, real-world social networks.

Abstract

Graph-based social recommendation systems have shown significant promise in enhancing recommendation performance, particularly in addressing the issue of data sparsity in user behaviors. Typically, these systems leverage Graph Neural Networks (GNNs) to capture user preferences by incorporating high-order social influences from observed social networks. However, existing graph-based social recommendations often overlook the fact that social networks are inherently noisy, containing task-irrelevant relationships that can hinder accurate user preference learning. The removal of these redundant social relations is crucial, yet it remains challenging due to the lack of ground truth. In this paper, we approach the social denoising problem from the perspective of graph invariant learning and propose a novel method, Social Graph Invariant Learning(SGIL). Specifically,SGIL aims to uncover stable user preferences within the input social graph, thereby enhancing the robustness of graph-based social recommendation systems. To achieve this goal, SGIL first simulates multiple noisy social environments through graph generators. It then seeks to learn environment-invariant user preferences by minimizing invariant risk across these environments. To further promote diversity in the generated social environments, we employ an adversarial training strategy to simulate more potential social noisy distributions. Extensive experimental results demonstrate the effectiveness of the proposed SGIL. The code is available at https://github.com/yimutianyang/SIGIR2025-SGIL.

Figures (5)

• Figure 1: Illustration of the existing filtering-based social denoising methods and our proposed invariance-based method. Instead of filtering the noisy part of the input social graph, we encourage learning the invariant yet sufficient user preferences across environments.
• Figure 2: Overview of our proposed SGIL framework, consists of two elaborated modules. (1) Environment Simulation: aiming to generate multiple social environments; (2) Invariant Preference Learning: targeting to achieve the invariant user preference across environments.
• Figure 3: Recommendation performance (NDCG@20) under different degrees of noise.
• Figure 4: Recommendation performance (NDCG@20) across different sparsity users.
• Figure 5: Impact of the number of environments $K$ and invariance coefficient $\beta$.