k-Anonymous microaggregation with preservation of statistical dependence

详细信息    查看全文

• 作者：David Rebollo-Monedero ; ^a ; ^{david.rebollo@entel.upc.edu" class="auth_mail" title="E-mail the corresponding author} ; ^{davidrm4@gmail.com" class="auth_mail" title="E-mail the corresponding author} ; Jordi Forné ; ^a ; ^{jforne@entel.upc.edu" class="auth_mail" title="E-mail the corresponding author} ; Miguel Soriano^a ; ^b ; ^{soriano@entel.upc.edu" class="auth_mail" title="E-mail the corresponding author} ; Jordi Puiggalí ; Allepuz^c ; ^{jordi.puiggali@scytl.com" class="auth_mail" title="E-mail the corresponding author}
• 关键词：k-Anonymity ; Microaggregation ; Statistical disclosure control ; Statistical dependence ; Predictability
• 刊名：Information Sciences
• 出版年：2016
• 出版时间：10 May 2016
• 年：2016
• 卷：342
• 期：Complete
• 页码：1-23
• 全文大小：1714 K

文摘

k-Anonymous microaggregation emerges as an essential building block in statistical disclosure control a field concerning the postprocessing of the demographic portion of surveys containing sensitive information in order to safeguard the anonymity of the respondents. Traditionally this form of microaggregation has been formulated to characterize both the privacy attained and the inherent information loss due to the aggregation of quasi-identifiers which may otherwise be exploited to reidentify the individuals to which a record in a published database refers. Because the ulterior purposes of such databases involve the analysis of the statistical dependence between demographic attributes and sensitive data we must articulate mechanisms to enable the preservation of the statistical dependence between quasi-identifiers and confidential attributes beyond the mere degradation of the quasi-identifiers alone.

This work addresses the problem of k-anonymous microaggregation with preservation of statistical dependence in a formal systematic manner modeling statistical dependence as predictability of the confidential attributes from the perturbed quasi-identifiers. We proceed by introducing a second mean squared error term in a combined Lagrangian cost that enables us to regulate the trade-off between quasi-identifier distortion and the confidential-attribute predictability. A Lagrangian multiplier enables us to gracefully weigh the importance of each of the two competing objectives.