基于输入区别本地差分隐私的键值数据收集机制OA

Local differential privacy(LDP)provides quantifiable privacy preservation for data collection without relying on a trusted third party.In real-world scenarios,key-value data,as typical non-relational data,usually exhibits diverse sensitivities across different keys.However,traditional LDP frameworks only enforce the same level of privacy protection for all key-value data.To address this limitation,this pa-per proposes a key-value data collection mechanism based on input discriminative local differential pri-vacy(ID-LDP)to accommodate personalized privacy requirements for different data items.First,we present the formal definition of ID-LDP and clarify its differences from conventional LDP.Second,we for-mulate the target optimization problem and identify the corresponding challenges.Third,we design a lo-cal perturbation mechanism and a server aggregation mechanism,and derive a method to compute utility-optimized perturbation probabilities.Finally,theoretical analysis and extensive experimental results dem-onstrate the superiority of the proposed mechanism over traditional approaches.