Scholarlink

<p>Graph Contrastive Learning (GCL) has made significant progress in recommendation systems, as they can explore high-order interaction relations between users and items, and provide better performance than traditional collaborative filtering algorithms in condition of sparse interactive data. However, these methods become fragile against carefully designed adversarial attacks and cannot maintain the reliability of recommendations. This paper proposes <i>Graph Contrastive learning adversarial Defense algorithm for Recommendation</i> (GCDR) to address this challenge. By integrating Random Perturbation Augmentation, Graph Filtering, and Parameters Initialization strategies, GCDR enhances robustness of recommendation systems under adversarial attacks. We conduct experimental evaluations on multiple benchmark datasets, and the results show that GCDR can significantly improve recommendation performance compared to existing baselines under adversarial attacks. This work not only reveals the security vulnerabilities of GCL, but also provides an effective technical approach for building more robust recommendation systems. Supplemental materials including code and data are available at <a href="https://github.com/DengBW-1998/GCDR">https://github.com/DengBW-1998/GCDR</a>.</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

GCDR: Graph Contrastive learning adversarial Defense algorithm for Recommendation

  • 期刊论文

  • 发表日期：

    2025年10月21日

查看全文
  • Bowen Deng,
  • Lin Feng,
  • Fancheng Yang,
  • Siwen Li,
  • Shuo Qin

摘要

Graph Contrastive Learning (GCL) has made significant progress in recommendation systems, as they can explore high-order interaction relations between users and items, and provide better performance than traditional collaborative filtering algorithms in condition of sparse interactive data. However, these methods become fragile against carefully designed adversarial attacks and cannot maintain the reliability of recommendations. This paper proposes Graph Contrastive learning adversarial Defense algorithm for Recommendation (GCDR) to address this challenge. By integrating Random Perturbation Augmentation, Graph Filtering, and Parameters Initialization strategies, GCDR enhances robustness of recommendation systems under adversarial attacks. We conduct experimental evaluations on multiple benchmark datasets, and the results show that GCDR can significantly improve recommendation performance compared to existing baselines under adversarial attacks. This work not only reveals the security vulnerabilities of GCL, but also provides an effective technical approach for building more robust recommendation systems. Supplemental materials including code and data are available at https://github.com/DengBW-1998/GCDR.