| 本文已被：浏览 5446次 下载 8448次 |
码上扫一扫！ |
|
|
| 图数据发布隐私保护的聚类匿名方法 |
|
姜火文1,2,3, 占清华4, 刘文娟1,3, 马海英5
|
|
1.同济大学 计算机科学与技术系, 上海 200092;2.江西科技师范大学 数学与计算机科学学院, 江西 南昌 330038;3.嵌入式系统与服务计算教育部重点实验室(同济大学), 上海 200092;4.江西科技学院 信息工程学院, 江西 南昌 330098;5.南通大学 计算机科学与技术学院, 江苏 南通 226019
|
|
| 摘要: |
| 社交网络中积累的海量信息构成一类图大数据，为防范隐私泄露，一般在发布此类数据时需要做匿名化处理.针对现有匿名方案难以防范同时以结构和属性信息为背景知识的攻击的不足，研究一种基于节点连接结构和属性值的属性图聚类匿名化方法，利用属性图表示社交网络数据，综合根据节点间的结构和属性相似度，将图中所有节点聚类成一些包含节点个数不小于k的超点，特别针对各超点进行匿名化处理.该方法中，超点的子图隐匿和属性概化可以分别防范一切基于结构和属性背景知识的识别攻击.另外，聚类过程平衡了节点间的连接紧密性和属性值相近性，有利于减小结构和属性的总体信息损失值，较好地维持数据的可用性.实验结果表明了该方法在实现算法功能和减少信息损失方面的有效性. |
| 关键词: 社交网络 隐私保护 聚类匿名 属性图 数据发布 |
| DOI：10.13328/j.cnki.jos.005178 |
| 分类号: |
| 基金项目:国家自然科学基金（61762044，71561013，61402244）；江西科技师范大学重点科研项目（2016XJZD002） |
|
| Clustering-Anonymity Approach for Privacy Preservation of Graph Data-Publishing |
|
JIANG Huo-Wen1,2,3, ZHAN Qing-Hua4, LIU Wen-Juan1,3, MA Hai-Ying5
|
|
1.Department of Computer Science and Technology, Tongji University, Shanghai 200092, China;2.School of Mathematics and Computer Science, Jiangxi Science and Technology Normal University, Nanchang 330038, China;3.Embedded System and Service Computing Key Laboratory of Ministry of Education(Tongji University), Shanghai 200092, China;4.College of Information Engineering, Jiangxi University of Technology, Nanchang 330098, China;5.College of Computer Science and Technology, Nantong University, Nantong 226019, China
|
| Abstract: |
| A huge amount of information in social network has accumulated into a kind of big graph data. Generally, to prevent privacy leakage, the data to be published need to be anonymized. Most of the existing anonymization scheme cannot prevent such attacks by background knowledge of both structure and attribute information among nodes. To address the issue, this investigation proposes a clustering-anonymization method for attribute-graph based on link edges and attributes value among nodes. Firstly, the data in the social network is represented by attribute graph. Then all the nodes of this attribute graph are clustered into certain super-nodes according to structural and attribute similarity between two nodes, each of which contains no less than k nodes. Finally, all the super-nodes are anonymized. In this method, the structure masking and attribute generalization for every super-nodes can respectively prevent all the recognition attacks by background knowledge of goals' linkages and attribute information. In addition, it balances the closeness of links among nodes and proximity of attributes value during clustering, therefore can reduce the total loss of information triggered by masking and generalization to maintain the availability of these graph data. Experiment results also demonstrate the approach achieves great algorithm performance and reduces information loss remarkably. |
| Key words: social network privacy preservation clustering-anonymity attribute graph data-publishing |
