摘要
针对现有的差分隐私聚类算法仅局限于实型数据的问题,提出一种基于混合型位置大数据的差分隐私聚类算法DPKD。利用KD-medoids降维聚类算法对混合型位置大数据进行预处理,提取位置信息记录,采用邻近搜索找出聚类中心点,划分为k个聚类簇,添加Laplace噪声使其满足差分隐私,通过查询函数返回待发布的数据记录;分析DPKD算法数据查询误差高的问题,对初始中心点优化选择,提出一种改进的Op-DPKD算法。性能评估结果表明,Op-DPKD算法解决了混合型位置大数据的隐私保护问题,提升了聚类效果,保证了混合型位置大数据的可用性。
Aiming at the problem that the existing differential privacy clustering algorithm is limited to real data,a differential privacy clustering algorithm based on mixed location big data,named as DPKD,was proposed.The KD-medoids dimension reduction clustering algorithm was used to preprocess the mixed location big data and extract the location information recorded from the mixed data set.The proximity search strategy was adopted to find the center point of clustering,which was divided into k clusters,and Laplace noise was added to satisfy the differential privacy protection mechanism.The data recorded to be published were returned through the query function.The problem of higher data query error of DPKD algorithm was analyzed,an improved algorithm named as Op-DPKD was proposed for the optimization of initial center point.The performance evaluation results show that the Op-DPKD algorithm solves the privacy protection problem of mixed location big data and improves the clustering effects,while ensuring the availability of mixed location big data.
作者
张建坤
禹思敏
ZHANG Jian-kun;YU Si-min(School of Automation,Guangdong University of Technology,Guangzhou 510006,China)
出处
《计算机工程与设计》
北大核心
2019年第9期2451-2455,2493,共6页
Computer Engineering and Design
基金
国家自然科学基金重点基金项目(61532020)
国家自然科学基金面上基金项目(61671161)
关键词
混合型位置大数据
差分隐私
聚类算法
隐私保护
数据预处理
mixed location big data
differential privacy
clustering algorithm
privacy protection
data preprocessing
