基于差分隐私的轨迹模式挖掘算法

Trajectory pattern mining with differential privacy

针对现有基于差分隐私的频繁轨迹模式挖掘算法全局敏感度过高、挖掘结果可用性较低的问题,提出一种基于前缀序列格和轨迹截断的差分隐私下频繁轨迹模式挖掘算法——LTPM。该算法首先利用自适应的方法获得最优截断长度,然后采用一种动态规划的策略对原始数据库进行截断处理,在此基础上,利用等价关系构建前缀序列格,并挖掘频繁轨迹模式。理论分析表明LTPM算法满足ε-差分隐私;实验结果表明,LTPM算法的准确率(TPR)和平均相对误差(ARE)明显优于N-gram和Prefix算法,能有效提高挖掘结果的可用性。

To address the problems of high global query sensitivity and low utility of mining results in the existing works, a Lattice-Trajectory Pattern Mining （LTPM） algorithm based on prefix sequence lattice and trajectory truncation was proposed for mining sequential patterns with differential privacy. An adaptive method was employed to obtain the optimal truncation length, and a dynamic programming strategy was used to truncate the original database. Based on the truncated database, the equivalent relation was used to construct the prefix sequence lattice for mining trajectory patterns. Theoretical analysis shows that LTPM satisfies c-differential privacy. The experimental results show that the True Postive Rate （TPR） and Average Relative Error （ARE） of LTPM are better than those of N-gram and Prefix algorithms, which verifies that LTPM can effectively improve the utility of the mining results.