A Relative Privacy Model for Effective Privacy Preservation in Transactional Data

Abstract

Transactional data such as shopping logs, web search queries and medical notes present enormous opportunities for knowledge discovery through data mining. When such data is published for knowledge discovery, privacy disclosure risks arise, making privacy preserving publication a fundamental requirement. However, existing publication mechanisms do not fully prevent an adversary from making an inference about the intended victim. While some solutions to this problem exist for the publication of relational data, they are not transferable to the publication of transactional data due to the difference in data models. This work aims to prevent inference attacks in the publication of transactional data by proposing a relative privacy metric that ensures that the knowledge gain of an adversary about any individual from the published data is bound to the general public knowledge. We then propose a publication mechanism Anony, which satisfies the proposed privacy metric without having to use excessively large cluster sizes. Finally, we evaluate our publication mechanism using two benchmark datasets and the results demonstrate that the proposed mechanism is effective and efficient.