基于事务分离的差分隐私频繁项集挖掘方法

Differential privacy frequent itemset mining method based on transaction separation

频繁项集挖掘在加入差分隐私后将带来敏感度过高、噪声过大、数据可用性较差的问题。为了解决这些问题,提出了基于事务分离的差分隐私频繁项集挖掘方法。利用指数机制对事务最大限制长度进行筛选,将长事务分离成为多个短事务,以此降低全局敏感度并避免截断误差的产生。在数据挖掘过程中,采用Apriori算法挖掘频繁项集,利用双阈值进行项集判断以及修正支持度,减小传输误差的产生和噪音。实验结果表明,该方法满足差分隐私的要求,可有效提高数据可用性。

After adding differential privacy,frequent itemset mining will bring problems such as high sensitivity,excessive noise and poor data usability.In order to solve these problems,a transaction separation based differential privacy frequent itemsets mining method is proposed.Using the index mechanism to filter the maximum transaction length and separate the long transaction into multiple short transactions to reduce the global sensitivity and avoid the truncation error.Firstly,the Apriori algorithm is used to mine frequent itemsets.Secondly,the dual threshold is used to judge itemsets and correct support,so as to reduce transmission error and noise.Experimental results show that this method meets the requirements of differential privacy and improves the data availability effectively.