A privacy preserving platform for MapReduce

Abstract

Big data applications typically require a large number of clusters, running in parallel, to process data fast and more efficiently. This is typically controlled and managed by MapReduce. In MapReduce operations, Mappers transform input original key/value pairs to a set of intermediate key/value pairs while Reducers aggregate a set of intermediate values, compute and write to the output. The output however can bring serious privacy concerns. Firstly, the output can directly leak sensitive information because it contains the global view of the final computation. Secondly, the output can also indirectly leak information via composite attacks where the adversary can link it with public information published via different sources such as Facebook or Twitter. To address such privacy concerns, we propose a privacy preserving platform which can prevent privacy leakage in MapReduce. Our platform can be plugged into the Reduce phase to sanitize the final output in such a way that the privacy is preserved while it yet provides a high data utility. We demonstrate the feasibility of our platform by providing empirical studies and highlights that our proposal can be used for real life applications.