Enhancing the Utility of Generalization for Privacy Preserving Re publication of Dynamic Datasets

Abstract

Anonymized publication on static micro data can be achieved with heavy information loss by Generalization. An enhanced utility of Generalization known as Angelization produces the same level of anonymization but with minimal information loss. In reality, there may be a need to publish another version of micro data, after insertions and deletions. Anonymization is applicable to any generalization principles like k-Anonymity, l-diversity and t-closeness. Incremental m-invariance with Angelization preserves privacy in re-publication of dynamic micro data after insertions and deletions. Mondrian algorithm is used for the partitioning in Angelization. m-invariance also supports publication of marginals from the generalized micro data. KL-divergence is employed for quantifying the discrepancy of two distributions. The reconstruction error will be measured as the KL-divergence between the reconstructed distribution and the original distribution. Data reconstruction error is minimal in m-invariance with enhanced utility of Generalization. General Terms Retrieval Model, Information Search and Retrieval.