Scholarlink

Privacy Preserving Record Linkage is the task of identifying the same real-world entities (usually humans) in databases originating from different dataholders, without revealing to any of them any other information apart from their matching records. In this paper, we focus on Privacy-Preserving Record Matching, the stage of Privacy-Preserving Record Linkage where records are checked for matching between data owners and we present FedeM: Federated Learning-based Privacy-Preserving Record Matching. FedeM is generic, relying on Federated Learning, without requiring a Linkage Unit, achieving high matching quality. Using Support Vector Machines, FedeM performs equivalently to a non-Federated Learning SVM-based setup for plain-text record matching, with only 1% decrease in Precision and 5% decrease in Recall.

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

FedeM: Federated Learning-Based Privacy-Preserving Record Matching

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Michail Zervas,
  • Alexandros Karakasidis

摘要

Privacy Preserving Record Linkage is the task of identifying the same real-world entities (usually humans) in databases originating from different dataholders, without revealing to any of them any other information apart from their matching records. In this paper, we focus on Privacy-Preserving Record Matching, the stage of Privacy-Preserving Record Linkage where records are checked for matching between data owners and we present FedeM: Federated Learning-based Privacy-Preserving Record Matching. FedeM is generic, relying on Federated Learning, without requiring a Linkage Unit, achieving high matching quality. Using Support Vector Machines, FedeM performs equivalently to a non-Federated Learning SVM-based setup for plain-text record matching, with only 1% decrease in Precision and 5% decrease in Recall.