Abstract:

In the big data environment, non-interactive differential privacy can not accurately provide and deal with a large number of queries. A privacy protection data query model based on maximum information coefficient and machine learning is proposed. Firstly, the data with low correlation is selected as the training sample set by using the maximum information coefficient of the original data set, and then combined with the parallel combination property of differential privacy to obtain the privacy-protected training sample set. Finally, the linear regression algorithm is used to train the sample. The differential privacy protection prediction model answers the current and a large number of unknown queries. The experimental results show that the proposed model improves the efficiency of query processing while improving the utility of published data.

Key words: differential privacy, maximum information coefficient, privacy protection, range query

摘要：

针对大数据环境下，非交互式差分隐私无法准确提供及处理大量范围查询的问题，提出一种基于最大信息系数与机器学习的隐私保护数据查询模型。对原始数据集采用最大信息系数选出相关性低的数据作为训练样本集，然后结合差分隐私的并行组合性质对其进行分块划分得到隐私保护的训练样本集，最后应用线性回归算法训练样本集得到差分隐私保护预测模型，该模型隐私保护的方式回答当前提交和大量未知的查询。实验结果表明，所提出的模型在提升发布数据效用性的同时，也提高了查询处理的效率。

关键词: 差分隐私, 最大信息系数, 隐私保护, 范围查询