多源用户画像差分隐私保护机制研究OA

To address privacy leakage risks in multi-source user profiling analysis,this paper proposes an end-to-end privacy protection framework based on differential privacy.The framework aggregates data at the user level,employs irreversible pseudonymization for identity protection,and integrates a persistent privacy budget management mechanism along with a pluggable extensible query engine.Experiments conducted on a real-world consumption dataset containing 8,636 users show that for the task of sum aggregation on consumption amounts,with a privacy budget set to 2.0,the average relative error is only 0.05%.In contrast,under the same scenario,the k-anonymity scheme(k=50)leads to a 0.4%data suppression rate and a 0.39%deterministic bias.The proposed framework enables secure statistical analysis of user profiling data,ensures data integrity,and provides a quantifiable privacy-utility trade-off superior to that of the k-anonymity approach.