A Weighted Context Graph Model for Fast Data Leak Detection

Data leakage prevention (DLP) uses a series of techniques to detect and prevent the sensitive data leakage caused by insider threat. Current detection methods either fail to achieve high accuracy toward transformed data or fail to reduce computational complexity. To ensure high detection accuracy and reduce computational complexity, we propose a Weighted Context Graph Model (WCGM) in this paper. The main goal of WCGM is three folds. First, the weighted context graph is proposed to build the contextual relation of data, based on which sub-graph matching method is used to calculate similarity features between tested data and pre-defined template. Second, machine learning algorithms are used to classify the tested data based on the similarity features of its context graphs. Third, privacy- preserving graph masking method is proposed to protect the data privacy of data holders. Extensive simulation results show that the proposed WCGM is able to achieve significant enhancement in terms of running time and accuracy.