面向大规模数据收集与机器学习的隐私保护技术

近年来，随着5G移动通信技术的迅速发展和商用部署，各种移动应用对移动用户数据的大规模收集更为便捷，随之而来的移动用户隐私保护问题也逐渐暴露出来。基于此，本项目以移动用户隐私保护和数据安全共享问题为切入点，提出了针对移动用户的隐私分析和保护框架，具体包括移动用户隐私量化分析、针对不同数据类型的移动用户隐私保护方法和基于透明的安全数据共享机制。研究成果将用于搭建移动用户场景下的隐私保护原型系统，并以移动通信领域为应用示范，以验证所提出的保护方法和共享机制在真实数据上的有效性和高效性。通过本项目研究可以为移动用户隐私保护和数据安全共享的进一步研究提供理论方法、技术支撑与新的思路。

项目题目

² 国家自然科学基金专项项目“移动用户隐私保护与数据安全共享理论与方法”（ 6194100069 ），2020年11月01日 - 2021年12月31日

项目说明

图1 数据隐私保护与安全共享框架

本项目从5G应用的角度出发，基于图1提出的数据隐私保护与安全共享框架，将主要研究内容概括为：（1）基于移动用户的隐私量化与分析研究，具体包括：研究如何识别和分析移动应用程序的数据收集情况，研究如何评估用户在使用移动应用程序时面临的隐私风险，研究如何设计有效的移动应用程序隐私风险预警机制；（2）基于本地化差分隐私的移动用户数据保护方法研究，具体包括：研究针对键值对数据的本地化差分隐私保护技术，研究针对图数据的本地化差分隐私保护技术，研究针对流数据的本地化差分隐私保护技术；（3）基于区块链的数据安全共享研究，具体包括：研究支持溯源问责的数据共享流通技术，研究可验证的分布式数据集共享技术，研究基于数据共享的、可验证的分布式机器学习。

项目工作

· PrivKV: Key-Value Data Collection with Local Differential Privacy

Local differential privacy (LDP), where each user perturbs her data locally before sending to an untrusted data collector, is a new and promising technique for privacy-preserving distributed data collection. To the best of our knowledge, there is no existing LDP work on key-value data, which is an extremely popular NoSQL data model and the generalized form of set-valued and numerical data. In this paper, we study this problem of frequency and mean estimation on key-value data by first designing a three methods, namely PrivKV, PrivKVM and PrivKVM+.

· AppPrivacy: Analyzing Data Collection and Privacy Leakage from Mobile Apps

While collecting legitimate usage data, many mobile applications (apps) have reportedly posed privacy threats to their hosted mobile devices and individuals, who are, unfortunately, unaware of data leaks and measures to protect themselves against these leaks. In this poster, we present a system that analyzes the two sides of mobile application ecosystem - data collection and privacy risk. The system consists of three main modules that correspond to mobile apps, users and service providers, respectively. To the best of our knowledge, this is the first work to evaluate privacy risk by analyzing data collection and privacy leakage from mobile apps.

· Protecting Location Privacy against Location-Dependent Attacks in Mobile Services

Most of the existing k-anonymity location cloaking algorithms are concerned with snapshot user locations only and cannot effectively prevent location-dependent attacks when users’ locations are continuously updated. Therefore, adopting both the location k-anonymity and cloaking granularity as privacy metrics, we propose a new incremental clique-based cloaking algorithm, called ICliqueCloak, to defend against location-dependent attacks. The efficiency and effectiveness of the proposed ICliqueCloak algorithm are validated by a series of carefully designed experiments. And, the experimental results also show that the price paid for defending against location-dependent attacks is small.

· 中国隐私风险指数分析报告

本报告对2019年度使用移动设备的用户(以下报告中简称移动用户)个人数据被收集情况进行调研分析，从移动场景下两大数据主体:数据拥有者（移动用户）、数据收集者（App开发者）角度入手，设计隐私风险量化模型定量并制定中国隐私风险指数体系，从而揭示隐私风险各维度特征。该报告以地域分层抽样得到的3000万真实用户为实验样本，分析移动用户App使用状况和隐私风险成因，通过数据收集者隐私风险指数（单App数据收集者、多App数据收集者）、数据拥有者隐私风险指数（区域隐私风险指数、人群隐私风险指数、行为隐私风险指数）讨论3000万用户的隐私风险群体特征。

项目成果

期刊论文

(1) 朱敏杰; 叶青青; 孟小峰; 杨鑫; 基于权限的移动应用程序隐私风险量化, 中国科学 : 信息科学, 2021, 51(7)：1100-1115. 第一标注

(2) 王雷霞; 孟小峰; ESA:一种新型的隐私保护框架, 计算机研究与发展, 2022, 59(1)：144-171. 第一标注

(3) 张祎; 孟小峰; InterTris:三元交互的领域知识图谱表示学习, 计算机学报, 2021, 44(8)：1535-1548. 第二标注

会议论文

(1) Qingqing Ye; Haibo Hu; Ninghui Li; Xiaofeng Meng; Huadi Zheng; Haotian Yan; Beyond Value Perturbation: Local Differential Privacy in the Temporal Setting, IEEE International Conference on Computer Communications, Virtual Conference, 2020-7-6至2020-7-9. 第一标注

(2) Qingqing Ye; Haibo Hu; Man Ho Au; Xiaofeng Meng; Xiaokui Xiao; Towards Locally Differentially Private Generic Graph Metric Estimation, International Conference on Data Engineering, Dallas, Texas, US, 2020-4-20至2020-4-24. 第二标注

(3) Jinfei Liu; Jian Lou; Junxu Liu; Li Xiong; Jian Pei; Jimeng Sun; Dealer: An End-to-End Model Marketplace with Differential Privacy, Proceedings of the VLDB Endowment, Copenhagen, Denmark, 2021-8-16至2021-8-20. 第七标注

(4) Qiongqiong Lin; Jiayao Zhang; Jinfei Liu; Kui Ren; Jian Lou; Junxu Liu; Li Xiong; Jian Pei; Jimeng Sun ; Demonstration of Dealer: An End-to-End Model Marketplace with Differential Privacy, Proceedings of the VLDB Endowment, Copenhagen, Denmark, 2021-8-16至2021-8-20. 第七标注

(5) Xinle Wu; Lei Wang; Shuo Wang; Xiaofeng Meng; Linfeng Li; Haitao Huang; Xiaohong Zhang; Jun Yan; A Unified Adversarial Learning Framework for Semi-supervised Multi-target Domain Adaptation, International Conference on Database Systems for Advanced Applications, Jeju, South Korea, 2020-9-24至2020-9-27. 第四标注