Anonymizing Transaction Data by Integrating Suppression and Generalization

Abstract

Privacy protection in publishing transaction data is an important problem. A key feature of transaction data is the extreme sparsity, which renders any single technique ineffective in anonymizing such data. Among recent works, some incur high information loss, some result in data hard to interpret, and some suffer from performance drawbacks. This paper proposes to integrate generalization and suppression to reduce information loss. However, the integration is non-trivial. We propose novel techniques to address the efficiency and scalability challenges. Extensive experiments on real world databases show that this approach outperforms the state-of-the-art methods, including global generalization, local generalization, and total suppression. In addition, transaction data anonymized by this approach can be analyzed by standard data mining tools, a property that local generalization fails to provide.