Sensitivity-Based Anonymization of Big Data

摘要

Data Analytics is widely used as a means of extracting useful information from available data. It is only natural that it is increasingly adapted for processing big data. The rapidly growing demand for big data analytics has several undesirable side-effects. Perhaps, the most significant of those relates to increased risks for data disclosure and privacy violations. Data anonymization can provide promising solutions for minimizing such risks. In this paper, we discuss some of the specific requirements of the anonymization process when dealing with big data. We show that in general, information loss is the result of avoidable generalization of similar or equivalent data. Using these analyses, we propose a novel framework for data anonymization, which expands the k-anonymity properties and concepts and takes the data class values and the sensitivity of data into account. As such, the proposed process can utilize a bottom-up approach, in contrast to most other anonymization methods. The top-down approaches usually generalize all records, the equivalent and the non-equivalent ones. Ours is more methodical, as it avoids the generalization of the equivalent records. With the inclusion of sensitivity levels, we demonstrate that our framework can reduce the iteration steps and the time required to finalize the anonymization, and therefore enhance the overall efficiency of the process.