Boost Social Recommendation via Adaptive Denoising Network

Xinran Chen, Chaobo He, Quanlong Guan

Abstract: Social recommendation aims to integrate social relationships to improve the performance of recommendation, and has attracted increasing attention in the field of recommendation system. Recently, Graph Neural Networks (GNNs) based methods for social recommendation are very competitive, but most of them overlook the fact that social relationships may have potential noises. Through the message passing mechanism of GNNs, these noises could be propagated and amplified, ultimately reducing the performance of recommendation. In view of this, we propose a novel GNN-based Adaptive Denoising Social Recommendation (ADSRec) method. It devises a denoising network, which can alleviate the impact of social relationships noises via the adaptive weight adjustment strategy. By further introducing the contrastive learning, the representations of users and items can be enhanced, leading to better recommendation results. Extensive experiments on three widely used datasets demonstrate the superiority of ADSRec over baselines.