Privacy preserving similarity joins using MapReduce

Abstract Similarity join is an essential operator in data processing, mining and analysis. However, it is resource intensive and time consuming, particularly when processing big data. There is also a need to ensure data confidentiality in similarity joins, as joining between two files may result in personal information disclosure. Based on these two considerations, this paper proposes a MapReduce-based similarity joins with differential privacy technology (hereafter, referred to as PSJoin). The proposed parallel algorithm is designed to achieve high efficiency, in terms of answering similarity join queries privately and effectively. Specifically, the use of PSJoin ensures the preservation of privacy during the similarity join process and in the published results. A new private global ordering approach is presented in this paper, which is designed to deal with potential disclosure issues during the process, and a differential private similarity function is provided for this algorithm. Findings from our evaluations using large-scale real-world datasets demonstrate that our method can effectively guarantee privacy with only minimal accuracy loss in similarity queries, while offering good scalability consistently.