福建电脑2026，Vol.42Issue(2)：7-15,9.DOI:10.16707/j.cnki.fjpc.2026.02.002
差分隐私HADPK-means++聚类算法
Differential Privacy HADPK-means++Clustering Algorithm
摘要
Abstract
To address the issue of cluster center deviation caused by noise accumulation during iterations in differentially private k-means clustering,which severely compromises clustering utility,this paper proposes a highly available differential privacy HADPK-means++algorithm.The method enhances initial center quality through reverse-order-sorting-based initial cluster center selection,optimizes sample partitioning by introducing a new metric combining intra-cluster and inter-cluster similarity,and corrects the noisy cluster centers using the transformation invariance of differential privacy to prevent them from deviating from the valid data range.Experiments on multiple real-world datasets,including Iris and Wine,demonstrate that under the same privacy budget,the proposed algorithm outperforms existing mainstream differentially private k-means algorithms in terms of both F-measure and Normalized Mutual Information(NMI).The HADPK-means++algorithm effectively suppresses cluster center deviation,thereby improving clustering utility and robustness.关键词
聚类/K-means算法/差分隐私Key words
Clustering/K-Means Algorithm/Differential Privacy分类
信息技术与安全科学引用本文复制引用
徐富国,李磊,陈涛..差分隐私HADPK-means++聚类算法[J].福建电脑,2026,42(2):7-15,9.基金项目
本文得到河南省哲学社会科学教育强省研究项目(No.2025JYQS1084)、商丘市哲学社会科学规划项目(商社规办[2025]3号)资助. （No.2025JYQS1084）
