Privacy-Preserving Frequency Estimation Method
Naoki Kawahara, Atsuko Miyaji, Tomoaki Mimoto · 2023
Differential privacy is one of the methods used to protect personal data by adding noise to personal data and its analysis results, so that the statistics obtained from the data can be safely disclosed. In this paper, we propose a new frequency estimation method by applying differential privacy to frequent itemset mining, which analyzes the frequencies of items owned by individuals in a dataset. Existing differential privacy mechanisms for frequency estimation have shown high usefulness when the privacy budget is large, i.e., when the privacy protection strength is low, but have been less useful when the privacy budget is small. Our proposed method uses a mechanism that does not change the distribution of the original dataset and anonymized one regardless of the size of the privacy budget, so that the proposed method is highly useful even when the privacy budget is small.