Value Extraction And Data Mining For Mobile Communication Data

The process of urbanization is formed by regular movements of human beings.It yields different functional zones in a city,such as residential zone and commercial zone.Consequently,there exists a close connection between the human mobility pattern and the city's zones.However,it is not easy to collect large-scale society-wide data that can precisely capture the underlying relations between the individual's movement and the regional functions.Hence,our knowledge for understanding the basic patterns of human mobility is still limited.In order to discover the functions of different regions in a city,we propose an affinity based method in this paper.The affinity is a recently introduced metric for measuring the correlation of two connecting node in a complex network.The proposed model groups different functional zones by measuring user's arrival/departure distribution via relative entropy.In addition to this,we also identify the intensity of each functional zone by taking kernel density estimation method.In the end,some experiments are conducted to evaluate our method with a large-scale real-life dataset,which consists of 3 million cellphone users'records from a period of one month.Our findings on the interaction between the mobility pattern and the regional functions can capture the city dynamics efficiently and provide a valuable reference for urban planners.Currently,the technology of region function discovering is not mature,yet.Many classic surveys use statistical method based on POI(position of interesting),which has a lot of limitation in it.This paper briefly reviews related works on the exploration of region function discovering using human behaviors.Finally,a new algorithm and feature are developed to tackle this problem.There are 3 main contributions in this paper:(1)An affinity measure based on KLD which involved in the context of each region was developed.Firstly,a novel feature,Kullback-Leibler divergence(KLD,also called relative entropy),is defined to measure the similarity of two regions.This feature,extracted from cell phone users'arriving/leaving dynamics,fits the research intuitively and reduces the data sparseness problem.In some cases,regions having different functions may have similar arriving/leaving distribution due to the noise and misdate.Considering such situations,we adopted affinity measure model to promote the accuracy.(2)According to the affinity distribution of each region we infer the territory of these functions by a graph clustering algorithm.Thus the regions in the different clusters have different functions.For the regions in the same cluster,the functionality intensity is identified by Kernel Density Estimation(KDE)which is commonly applied to the community network.(3)An advanced LDA model was developed.Unlike the LDA model,the prior distribution over topics,a,is a function observed POI feature,and is therefore specific to each distinct combination of the POI feature values.Our findings on the interaction between the mobility pattern and the regional functions can capture the city dynamics efficiently and provide a valuable reference for urban planners.