Important Position Recognition And Regional Function Discovery Based On Cellular Network Data

With the development of smart phones,more and more people are used to carry mobile phones anywhere and anytime.Meanwhile,more functions and sensors are integrated into smart phones.Plenty of records are produced when smart phones are used by users.Every time users use the service presented by Operators(such as voice calls,SMS,traffic data and so on),these records are called cellular network data and will be stored as logs in Operators' databases.Cellular network data contains the time,location and service type of each record,but the specific communication content is excluded.The biggest advantage of cellular network data is the large-scale users and large-scale space it covered.Although records are sparse in both time and space,it still provides sufficient information for behavior research.for voice calls,SMS or traffic data.Some records are stored locally in phones,while others may be stored in databases of Operators as logs.Records in Operator's databases are called cellular network data.In this paper,we propose a method to identify the important position of the user based on the cellular network data.The method does not rely on any prior knowledge,i.e.it is purely data-driven.Although,the Cellular network data is sparse,the method still works.First,we extract features of each location one visited from the his/her communication logs to characterize his/her behavior at those locations.Then,we analyze the behavior features with cluster method.From results of cluster,we find that the difference between behavior of various locations are very distinguish:behaviors are very sparse in most common locations.But in some special positions,there are a lot of communication behaviors.Moreover,the behavior of these special location differs from each other.The behavior patterns displayed in these locations are also different from each other.We call the location which has intensive communication behavior and contains distinguish behavior pattern the special location.Based on the analysis of the special location and the information from volunteers,we believe that the user's important location exists near the special location.In addition,classifiers are constructed with the truth value provided by volunteers to identify user's home and work location.90%of the predicted locations have a prediction error with 1600m.And we analyze the importance of behavior feature in the special locations(work and home location).The results show that from 0:00 am to 8:00 am,users are likely to be found at home,and there is not any obvious communication behavior at home.From 12:00 am to 20:00 pm,most users may be at work,and they tend to make more voice calls or SMS than at other location.Finally,we apply the constructed classifiers to all user in the network.So,we get the home and work locations of all users.With help of these locations,we estimate the distribution of residential and office function of Shanghai.The distribution of the work area in Shanghai is more concentrated than the living area,but the distribution of residential region and office region is basically the same.