基于事务截断的差分隐私频繁模式挖掘算法

Frequent Pattern Mining with Differential Privacy Based on Transaction Truncation

现有基于ε-差分隐私模型的频繁模式挖掘算法存在全局敏感度过高与挖掘结果可用性较低的不足.设计一个基于事务截断的差分隐私频繁模式挖掘算法.算法首先采用基于指数机制的事务截断思想,对长事务进行截断处理,以有效降低算法的全局敏感度,并在此基础上提出基于事务截断的差分隐私频繁模式挖掘算法,而后提出可用于扩充Apriori算法候选集的最小噪声支持度标准,以进一步提升挖掘结果的可用性.实验对本文算法的频繁模式挖掘结果与同类算法进行比较分析.实验结果表明,本文算法可在满足ε-差分隐私的前提下,保证挖掘结果具有较高的可用性.

Existing frequent pattern mining with the ε-differential privacy model has shortcoming of its greater global sensitivity and lower availability issues. We proposed a newfrequent pattern mining with differential privacy algorithm based on the transaction truncation. The algorithm firstly used the idea of transaction truncation based on exponential mechanism,truncating the long transaction to reduce the global sensitivity. Besides,we also proposed frequent pattern mining with differential privacy based on transaction truncation on this algorithm,and then we put forward the minimum noise support standard to expand the candidate set of the Apriori algorithm,which enhances the availability of data. Experiments compared and analyzed our algorithm results with similar algorithms. Experimental results showthat the algorithm can meet the ε-differential privacy and ensure high availability of mining results at the same time.