Scholarlink

<p>With the development of O2O e-commerce, POI recommendation services are increasingly important in integrating online and offline consumer scenarios. However, the growing complexity of consumer and merchant data brings POI recommendation dilemma in privacy protection and recommendation quality. Addressing these issues, we propose a privacy-enhanced personalized POI recommendation method (PEP-PRec) with federated learning framework. This framework includes three modules: (1) Data Privacy Enhancement Module, dynamically obfuscates social and check-in data on local devices. (2) Local Model Construction Module, uses a weighted matrix factorization algorithm to build localized user preference models incorporating multidimensional factors. (3) Recommendation Generation Module, aggregates and updates parameters from localized models via a central parameter server, creating an optimal global model that provides a Top-<i>N</i> recommendation list to users. Experimental results show that PEP-PRec outperforms advanced POI recommendation methods across multiple metrics, demonstrating its ability to enhance recommendation effectiveness while protecting user privacy.</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

Privacy-enhanced personalized POI recommendation with federated learning framework

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月8日

查看全文
  • Chunhua Ju,
  • Zhaohui Wang,
  • Chonghuan Xu,
  • Fuguang Bao

摘要

With the development of O2O e-commerce, POI recommendation services are increasingly important in integrating online and offline consumer scenarios. However, the growing complexity of consumer and merchant data brings POI recommendation dilemma in privacy protection and recommendation quality. Addressing these issues, we propose a privacy-enhanced personalized POI recommendation method (PEP-PRec) with federated learning framework. This framework includes three modules: (1) Data Privacy Enhancement Module, dynamically obfuscates social and check-in data on local devices. (2) Local Model Construction Module, uses a weighted matrix factorization algorithm to build localized user preference models incorporating multidimensional factors. (3) Recommendation Generation Module, aggregates and updates parameters from localized models via a central parameter server, creating an optimal global model that provides a Top-N recommendation list to users. Experimental results show that PEP-PRec outperforms advanced POI recommendation methods across multiple metrics, demonstrating its ability to enhance recommendation effectiveness while protecting user privacy.