Privacy preservation for re-publication data using probabilistic graph

Abstract

With the dynamic dataset could be changed by subjecting insert, update, and delete operations, at any time. Thus, the existing privacy models designed for protecting the static dataset could not cope with this case. The well-known privacy models for dynamic datasets, such as m-invariance and m-distinct are proposed to handle such kinds of the dataset. However, m-invariance cannot be used on fully dynamic re-publication datasets and m-distinct is a counting-based model that privacy violation could occur when we analyze by using the probabilistic graph. In this research, we propose an improvement for protecting privacy of fully dynamic re-publication dataset. We use the probabilistic graph in our calculation and propose a greedy algorithm that preserves privacy and does not produce much loss of information. It is calculated by using only probability paths from individual target that reduces the complexity of the calculation. From the experiment results, our proposed model can guarantee the maximum probabil- ity of inference for sensitive values and keep the loss of information not too high, about 1%/tuple/attribute. But it suits a probabilistic graph that does not has the high difference of probability of updating, about 0.0010 and several the number of releases, about not greater than 4 times. The results lead to the way for improvement that the grouping of quasi-identifier has to be considered. Nonetheless many points in our research can be improved in the future.