Data Dependent Itemset Mining Under Local Differential Privacy

Frequent itemset mining with Apriori-based methods is widely used to uncover patterns in large datasets. However, under the private setting of Local Differential Privacy (LDP), this task poses key challenges, such as domain scalability and user set-valued data. Prior works address these problems through fixed-size transaction reporting protocols but often rely on hard-coded parameters, overlooking complexities from differing dataset profiles. To address this limitation, we propose a data-dependent framework for identifying the top-k most frequent itemsets under LDP. Through initial metadata queries and the integration of an optimization problem, we significantly improve over prior work on both synthetic and real-world datasets.