Collaborative Graph-Based Mechanism for Distributed Big Data Leakage Prevention

Data leakage is a growing insider threat to data owners. Several studies have been done on data leakage prevention (DLP). In the era of big data, massive data has been generated constantly by various institutions. In many applications, multiple institutions may be interested in sharing data with each other to extract more value, without leaking private data. It is a new challenge for DLP in the big data scenario to train a global detection model over distributed big data sets without breaking the data privacy of each party. Moreover, as the forms of data become much complicated, the model should also be capable of tolerating data transformation. In this paper, we propose a Collaborative graph-based mechanism for Distributed big data Leakage Detection (CoDLD). CoDLD addresses the collaborative DLP problem in three aspects. First, it transfers the text detection problem into graph space. The local weighted graphs of data owners are iteratively constructed in turn. Second, it protects the privacy of each data owner by using graph masking on local weighted graphs. Third, it applies partition-based method on the graph to perform accurate matching detection efficiently. Experimental results show that our method can perform collaborative data leak detection over distributed data with high accuracy and efficiency.