两方参与的隐私保护协同过滤推荐研究  

Research on Privacy-Preserving Two-Party Collaborative Filtering Recommendation

作　　者：张锋[1,2] 孙雪冬[1] 常会友[1] 赵淦森[1] 

机构地区：[1]中山大学软件学院,广东广州510275 [2]中山大学广东省信息安全重点实验室,广东广州510275

出　　处：《电子学报》2009年第1期84-89,共6页Acta Electronica Sinica

基　　金：广东省自然科学基金重点项目(No.05100302);广东省信息安全技术重点实验室开放基金;中山大学青年教师科研启动基金(No.1131014)

摘　　要：隐私保护的协同过滤推荐研究致力于在确保高质、高效地产生推荐的同时有效地保护参与方的隐私.在数据分布存储,参与方大于2的情形,已有研究针对其核心任务——对指定项进行评分预测,以可交换的密码系统为主要技术,设计了一个隐私保护计算协议.但该协议不适用于参与方是2的情形.以安全比较计算和安全点积计算为基础安全设施,设计了一个协议,解决参与方是2的情况下对指定项进行评分预测的隐私保护问题,从而解决了隐私保护的两方协同计算问题.预测准确度与数据集中存放一样,证明了协议的正确性,并基于安全多方计算理论和模拟范例,证明其安全性,分析了时间复杂度和通信耗费.Privacy-preserving collaborative filtering aims at protecting participating parties＇ privacy while providing highquality recommendations efficiently. In the case of the number of the participating parties is greater than 2, a protocol, employing commutative encryption as its major privacy-preserving technique, has been devised to address the issue of rating a specific item in scenarios with distributed data storage, which is a key challenge in privacy-preserving collaborative filtering recommendation in that scenarios. However, the protocol does not work when the number of the participating parties is exactly 2. Employing secure comparison and secure dot product as its fundamental security infrastructure, we design a privacy-preserving two-party collaborative computing protocol to address the challenge. This protocol produces the same results as the traditional memory-based collaborative filtering recommender systems. Based on secure multi-party computation theory and simulation paradigm, the protocol＇ s security is proved. The protocol＇ s computation complexity and communication cost are examined as well.

关 键 词：隐私保护数据挖掘 安全多方计算 推荐系统 协同过滤 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]