Differential privacy histogram publishing method based on dynamic sliding window

Differential privacy has recently become a widely recognized strict privacy protection model of data release. Differential privacy histogram publishing can directly show the statistical data distribution under the premise of ensuring user privacy for data query, sharing, and analysis. The dynamic data release is a study with a wide range of current industry needs. However, the amount of data varies considerably over different periods. Unreasonable data processing will result in the risk of users’ information leakage and unavailability of the data. Therefore, we designed a differential privacy histogram publishing method based on the dynamic sliding window of LSTM (DPHP-DL), which can improve data availability on the premise of guaranteeing data privacy. DPHP-DL is integrated by DSW-LSTM and DPHK+. DSW-LSTM updates the size of sliding windows based on data value prediction via long short-term memory (LSTM) networks, which evenly divides the data stream into several windows. DPHK+ heuristically publishes non-isometric histograms based on k-mean++ clustering of automatically obtaining the optimal K, so as to achieve differential privacy histogram publishing of dynamic data. Extensive experiments on real-world dynamic datasets demonstrate the superior performance of the DPHP-DL.