UKVLDP: Utility-Optimized Local Differential Privacy Mechanism for Key–Value IoT Data Collection
Bin Wang, Chao Yang, Jianfeng Ma · IEEE Internet of Things Journal · 2025
Collecting key–value data from Internet of Things (IoT) devices is crucial for providing users with more accurate recommendation services and other applications. However, existing collection protocols for key–value data cannot provide differential privacy protection for data with varying sensitivities, which affects data utility. In this article, we introduce a novel utility-optimized local differential privacy definition for key–value data, UKVLDP. This definition enables key–value data of different sensitivity levels to be properly protected by appropriately relaxing the strict local differential privacy (LDP) definition, and adaptively perturbing them according to their different sensitivities, thereby improving the utility of the collected data while protecting the privacy of the key–value data. Based on UKVLDP, we propose a privacy-sensitivity-calibrated key–value data sampling algorithm for the privacy-preserving collection of key–value IoT data of different sensitivities, which solves the distortion problem of the random uniform sampling method when the key domain space is too large. In addition, we propose a key–value data collection protocol UKV-GRR that conforms to the UKVLDP definition, which achieves a better privacy-utility tradeoff by adaptively perturbing the sampled key–value data. We theoretically prove that UKVLDP can provide optimized privacy guarantees. Evaluations on simulated and real datasets demonstrate that UKV-GRR outperforms the state-of-the-art key–value generalized randomized response mechanism (PCKV-GRR), with an average enhancement of 31.6% in data utility. Our work offers a novel, practical solution for key–value IoT data collection mechanisms.