面向轨迹聚类的差分隐私保护方法

Differential privacy preserving method for trajectory clustering

针对目前的轨迹聚类隐私保护方法存在适用性较窄、可用性较低及难以在实际应用中实施的问题,提出了支持轨迹聚类的差分隐私保护方法.首先给出了典型轨迹聚类算法的通用框架模型及其差分隐私定义,然后根据定义设计满足差分隐私机制的二维拉普拉斯噪声,最后将直角坐标系中得到的噪声形式变换到极坐标系,并加入到原始轨迹点中以进行实际应用实现.实验结果表明：与当前的轨迹聚类隐私保护方法相比,本文算法具有更好的适用性和聚类效果.

As existing privacy preserving mechanisms for trajectory clustering are still faced with the problems of narrow applicability,low-level utility,which are difficult to imply in real scenarios,a differential privacy preserving mechanism was proposed to support trajectory clustering. Firstly,general framework model of typical trajectory clustering algorithms was given and the definition of differential privacy was introduced according to the framework. Then,the probability density function of two-dimensional Laplace noise satisfying the above definitions was derived. Finally, the noise from Cartesian coordinate system was transformed to Polar coordinate system to imply it efficiently. Experimental results show that compared with present methods,the proposed mechanism has general application and better cluster performance under the same preserving intensity.