Non-Centralized Distinct L-Diversity

This paper considers the non-centralized version of privacy preserving data publishing (PPDP), which refers to generating published tables from multiple non-centralized private tables owned by different data holders. Traditional solutions to PPDP on a single centralized dataset cannot be directly applied to this problem. Even if every published table satisfies a traditional privacy preserving requirement individually, an adversary who can collect multiple published tables may be able to deduce some private information that violates the satisfied requirement. Due to privacy reasons, the data holders cannot share information with each other to cooperate on the data publishing issues. In this paper, we propose non-centralizeddistinct l-diversity and an algorithm to generate published tables. Our algorithm does not rely on any communications between the data holders but only collects published tables released by other data holders. Experiments on real datasets are conducted to show that the algorithm is feasible to real applications.