摘要
挖掘位置数据中的用户行为规律是大数据时代的研究热点之一.现有研究主要关注于用户在某时刻出现在某地点的行为,对于用户从一个地点移动到另一个地点的动态行为研究较为空缺.提出一种挖掘位置数据中用户移动行为的算法可以发现用户的多个周期移动行为,描述用户在时空上的移动规律.首先,利用离散傅里叶变换和自相关系数检测用户移动行为的周期,在这一过程中,利用Apriori性质减少计算复杂度;而后提出用户移动行为的生成模型,估计用户的移动行为概率矩阵,考虑到观测数据的稀疏性,采用带全局限制的动态时间规整距离对不同时间段的行为进行聚类以发现用户的多个周期移动行为.最后,我们选取某市公共自行车系统收集的位置数据进行实证分析,结果表明,新方法能有效地挖掘用户的多个周期移动行为,进一步地,通过归纳可以得到用户群体在周期移动行为上的主要特征.
Mining the rule of users’ behaviors in location data is one of the research hotspots in the era of big data.Existing researches focus on the behavior of users appearing at a certain place at a certain moment,and the dynamic behavior research of users moving from one place to another is relatively empty.Proposing an algorithm to mine user’s moving behavior based on location data can discover users’ multiple periodic moving behaviors and describe users’ moving rules in time and space.Firstly,we use Discrete Fourier Transform and Autocorrelation Coefficient to detect the period of users’ moving behaviors.In this process,Apriori property is considered to reduce the computational complexity.Then,we propose a generative model and estimate the probability matrix of users’ moving behaviors.Taking into account the sparsity of users’ observation data,we cluster the behaviors of different time periods based on the dynamic time warping distance with global constraints to find the users’ multiple periodic moving behaviors.Finally,we select the location data collected by the public bicycle system in some city for empirical analysis.The results show that the new method can effectively mine users’ multiple periodic moving behaviors.Moreover,we summarize the main characteristics of the periodic moving behaviors for the users group.
作者
范一苇
吕晓玲
FAN Yi-wei;LU Xiao-ling(Center of Applied Statistics,Renmin University of China,Beijing 100872,China;Department of Statistics,Renmin University of China,Beijing 100872,China)
出处
《数学的实践与认识》
北大核心
2019年第14期181-190,共10页
Mathematics in Practice and Theory
基金
国家自然科学基金(61472475)
中央高校建设世界一流大学(学科)和特色发展引导专项资金
关键词
位置数据
移动行为
周期检测
聚类
动态时间规整
location data
moving behavior
period detection
clustering
dynamic time warping
