基于差分隐私的频繁序列模式挖掘算法

发布时间：2018-05-31 08:13

本文选题：频繁序列挖掘 + 差分隐私　；参考：《计算机应用》2017年02期

【摘要】：针对当数据集含有敏感信息时,直接发布频繁序列模式本身及其支持度计数都有可能泄露用户隐私信息的问题,提出一种满足差分隐私(DP)的频繁序列模式挖掘(DP-FSM)算法。该算法利用向下封闭性质生成候选序列模式集,基于智能截断方法从候选模式中挑选出频繁的序列模式,最后采用几何机制对所选出模式的真实支持度添加噪声进行扰动。另外,为了提高挖掘结果的可用性,设计了一个阈值修正的策略来减小挖掘过程中的截断误差和传播误差。理论分析证明了该算法满足ε-差分隐私。实验结果表明了该算法在拒真率(FNR)和相对支持度误差(RSE)两个指标上明显低于对比算法PFS2,有效地提高了挖掘结果的准确度。
[Abstract]:In order to solve the problem that when data sets contain sensitive information, it is possible to disclose user privacy information by directly publishing frequent sequence patterns and their support counts, a DP-FSM-based algorithm for frequent sequence pattern mining satisfying differential privacy DP) is proposed. The algorithm uses the property of downward closure to generate candidate sequence pattern sets, selects frequent sequential patterns from candidate patterns based on intelligent truncation method, and finally uses geometric mechanism to add noise to the true support of the selected patterns. In addition, in order to improve the usability of mining results, a threshold correction strategy is designed to reduce the truncation error and propagation error in the mining process. Theoretical analysis shows that the algorithm satisfies 蔚-difference privacy. The experimental results show that the proposed algorithm is obviously lower than the contrast algorithm PFS2 in two indexes of rejection rate (FNR) and relative support error (RSE), and the accuracy of mining results is improved effectively.
【作者单位】：东北大学计算机科学与工程学院;
【基金】：国家自然科学基金资助项目(61033007,61622202,61572119) 国家973计划项目(2012CB316201) 教育部中央高校基本科研业务费资助项目(N150402005)~~
【分类号】：TP311.13;TP309

【相似文献】