电子学报 ›› 2018, Vol. 46 ›› Issue (12): 3050-3059.DOI: 10.3969/j.issn.0372-2112.2018.12.032

• 学术论文 • 上一篇    下一篇

融合显/隐式信任协同过滤算法的差分隐私保护

鲜征征1,2, 李启良2, 黄晓宇3, 陆寄远1, 李磊2   

  1. 1. 广东金融学院互联网金融与信息工程学院, 广东广州 510521;
    2. 中山大学数据科学与计算机学院, 广东广州 510006;
    3. 华南理工大学经济与贸易学院, 广东广州 510006
  • 收稿日期:2017-06-13 修回日期:2018-04-22 出版日期:2018-12-25
    • 通讯作者:
    • 黄晓宇
    • 作者简介:
    • 鲜征征 女.1977年8月出生于四川省阆中市.博士,现为广东金融学院讲师,CCF会员.主要研究方向为数据挖掘、隐私保护等.E-mail:xianzhengzheng@126.com;李启良 男.1990年5月出生于广东省云浮市.硕士,现为华为技术有限公司集成服务部的软件工程师.主要研究方向为数据挖掘和隐私保护.E-mail:liqiliang90@163.com
    • 基金资助:
    • 广东省自然科学基金 (No.2017A030313391); 广东省科技计划 (No.2017A050501042,No.2016ZC0039,No.2017ZC0117); 广东省哲学社科 (No.GD15CGL05)

Differential Privacy Protection for Collaborative Filtering Algorithms with Explicit and Implicit Trust

XIAN Zheng-zheng1,2, LI Qi-liang2, HUANG Xiao-yu3, LU Ji-yuan1, LI Lei2   

  1. 1.School of Internet Finance and Information Engineering, Guangdong University of Finance, Guangzhou, Guangdong 510521, China;
    2.School of Data and Computer Science, Sun Yat-sen University, Guangzhou, Guangdong 510006, China;
    3.School of Economics and Commerce, South China University of Technology, Guangzhou, Guangdong 510006, China
  • Received:2017-06-13 Revised:2018-04-22 Online:2018-12-25 Published:2018-12-25
    • Corresponding author:
    • HUANG Xiao-yu
    • Supported by:
    • National Natural Science Foundation of Guangdong Province,  China (No.2017A030313391); Science and Technology Project of Guangdong Provicne (No.2017A050501042, No.2016ZC0039, No.2017ZC0117); Guangdong Philosophy and Social Science Foundation (No.GD15CGL05)

摘要: 融合显/隐式信任关系的社会化协同过滤算法TrustSVD在推荐系统中有广泛的应用,但该算法存在用户隐私泄漏的风险.基于背景知识的用户个人隐私信息推断是当前Internet用户隐私信息泄漏的巨大隐患之一,差分隐私作为一种能为保护对象提供严格的理论保证的隐私保护机制而备受关注.本文把差分隐私保护技术引入TrustSVD中,提出了具有隐私保护能力的新模型DPTrustSVD.理论分析和实验结果显示,DPTrustSVD不仅为用户的隐私信息提供了严格的理论保证,而且仍然保持了较高的预测准确率.

关键词: 社会化协同过滤, 个人隐私保护, 差分隐私, 矩阵分解, 信任关系, 隐式信任

Abstract: TrustSVD, a popular social collaborative filtering algorithm that incorporates both of the explicit and implicit trust information, has been widely used in recommender systems. However, there is a risk of disclosure of user privacy in TrustSVD. Privacy information inference based on background knowledge is one of the great hidden dangers of user's privacy disclosure. Differential privacy has attracted much attentiaon as a privacy protection mechanism that can provide a strict theoretical guarantee for protection objects. In this article, we propose DPTrustSVD, a novel collaborative filtering algorithm that applies Differential privacy to TrustSVD and has the ability of privacy preserving. Theoretical analysis and experimental results show that DPTrustSVD not only provides a strict theoretical guarantee for users' privacy information, but also maintains a high prediction accuracy.

Key words: social collaborative filtering, personal privacy preservation, differential privacy, matrix factorization, trust relationship, implicit trust

中图分类号: