Mining User's Mobility Patterns Based On Smartphone Location History

In recent years,intelligent terminals such as smart phones,PDAs and wearable devices integrate more and more sensors,and their processing performance is rapidly improved.These smart terminals are integrated with GPS sensors.Meanwhile,satellite positioning technology and mobile Internet technology develop fast,and people can use the intelligent terminal to more easily locate their mobile location,thus accumulated a large number of track data.How to analyze and study the information contained in these trajectory data becomes a hotspot in the field of mobile sensing,pervasive computing and trajectory data mining.In order to solve the problems of current research about user's mobility patterns based on trajectory data,such as mobility periodical parameters needing human intervention,periodicity discovering of time series data having imprecise result,and the mined mobility pattern being single or incomplete,this paper proposes PMPM(Periodic Mobility Pattern Mining)method based on heuristic data preprocessing combined with stay point mining,which method can detect the period of user's mobility trajectory data adaptively and mine user's periodic movement pattern based on period length.Firstly,the user's trajectory sequence is clustered to get the set of user's interest points.Then,the user's period length is detected for each point of interest points,and the frequent pattern mining algorithm is used to mine frequent mobility pattern and sequence mobility pattern.The feasibility and validity of the proposed method are proved by experiments on the open data set of the Geolife project of Microsoft Research Asia.Work in this paper is mainly reflected in the following three aspects:(1)A heuristic data preprocessing algorithm,VAPruning,is proposed to improve the efficiency of trajectory clustering by combining the algorithm of stay points mining.The data preprocessing algorithm proposed in this paper is effective for impurity removal and filtering of trajectory data,as well as effective for pruning trajectory data retained the points representing user's mobile characteristic in order to improve the efficiency of stay point mining even existing the problem of data imbalance,data redundancy and the existence of noise data in GPS trajectory data.(2)A time series distribution period detection model based on Fourier analysis is proposed.In this paper,we propose a two-layer model for discovering the period length of the time series.The cross-entropy verification based on the Fourier analysis results in improving the accuracy of the period detection.And solved the problem of periodic parameters in the mobility pattern mining being dependent on the empirical input.(3)This paper proposes a different classification of mobility patterns method based on movement sequence transformation.In this paper,we propose a method of movement sequence transformation for multi-dimensional GPS trajectory sequences including longitude,latitude,altitude,time stamp and so on.We use user's interest point set to replace the original trajectory data,converting the GPS trajectory sequences into digital tag sequences,and transforming multidimensional data for one-dimensional digital data.And we can mine mobility patterns based on digital sequences,reducing the efficiency requirements of mining algorithms,solving the problem of combinatorial explosion in high-dimensional data mining.