Scholarlink

<p>Location-based services (LBS) have significantly advanced the development of personalized location recommendation systems. However, two main challenges remain: safeguarding user privacy, as these systems rely on user data to provide tailored Point-of-Interest (POI) recommendations, and managing high communication costs due to the decentralized construction of neighbor sets in federated learning-based systems. To address these issues, this paper proposes a novel approach, called Category-Aware Privacy preserving Semi-Decentralized Bayesian Personalized Ranking for POI recommendation (CA-PSDBPR). This method integrates differential privacy to protect user POI data and encrypt gradient updates, thereby enhancing privacy protection. Additionally, clustering techniques are employed to group users into distinct clusters, enabling efficient intra-cluster communication, and thereby significantly reducing communication costs, which is ideal for federated learning scenarios. Our framework involves four key steps: encrypting user POI category vectors with differential privacy, using the <i>k</i>-means algorithm to cluster the encrypted vectors, updating local models using an improved BPR loss function, and exchanging gradient information within clusters with differential privacy protection. Experimental results on five real-world datasets indicate that CA-PSDBPR not only attains relatively high recommendation performance, but also enhances privacy protection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Category-aware privacy preserving semi-decentralized BPR for POI recommendation with user clustering and encryption

  • 期刊论文

  • 发表日期：

    2026年5月6日

查看全文
  • Bilian Chen,
  • Langcai Cao,
  • Renxu Wang,
  • Xiyang Lin,
  • Zichen Yang

摘要

Location-based services (LBS) have significantly advanced the development of personalized location recommendation systems. However, two main challenges remain: safeguarding user privacy, as these systems rely on user data to provide tailored Point-of-Interest (POI) recommendations, and managing high communication costs due to the decentralized construction of neighbor sets in federated learning-based systems. To address these issues, this paper proposes a novel approach, called Category-Aware Privacy preserving Semi-Decentralized Bayesian Personalized Ranking for POI recommendation (CA-PSDBPR). This method integrates differential privacy to protect user POI data and encrypt gradient updates, thereby enhancing privacy protection. Additionally, clustering techniques are employed to group users into distinct clusters, enabling efficient intra-cluster communication, and thereby significantly reducing communication costs, which is ideal for federated learning scenarios. Our framework involves four key steps: encrypting user POI category vectors with differential privacy, using the k-means algorithm to cluster the encrypted vectors, updating local models using an improved BPR loss function, and exchanging gradient information within clusters with differential privacy protection. Experimental results on five real-world datasets indicate that CA-PSDBPR not only attains relatively high recommendation performance, but also enhances privacy protection.