Differentially Private Frequent Itemset Mining from Smart Devices in Local Setting

Abstract

Frequent itemset mining has become an important approach of smart devices to upgrade service level for users, but comes with risks to privacy. And privacy leakage will result in serious consequence. Accordingly, it is highly desirable to mine frequent itemset while protecting users’ privacy. Moreover, users may not trust anyone else (including the miner) and are willing to share their information only if it has been perturbed appropriately before leaving their smart devices. Local differential privacy resolves this problem by only aggregating randomized itemsets from each user, with providing plausible deniability; meanwhile the miner can still obtain relatively accurate frequent patterns. Moreover users might have diverse privacy requirements on different items. These facts have led to the personalized differentially private frequent itemset mining, which preserves privacy with stochastic responses. Motivated by this, we propose a novel personalized local privacy preservation scheme for smart devices, which retains desirable accurate results while providing rigorous privacy guarantees.