آنالیز رفتارهای کاربر توسط کاوش مجموعه های داده شبکه ای بزرگ

ABSTRACT Analysis of user behaviors by mining large network data sets

Although technologies have been pushing our world and society to a smarter one, human behaviors in the society still keep some inherent features and complexities that are hard to explain. Understanding the intelligence of human behaviors in the real world has great significance in practical application, such as mobile network deployment, traffic engineering, urban planning and service recommendation. While studies on human behaviors were not new in social science, quantitative analyses were not common due to the lack of source of data. Thanks to the computers and networks as they can now give plenty of computational ways of collecting and analyzing data for social studies, which used to depend on surveys in traditional methodology. Thus in the new era of ‘‘Big Data’’, never before have researchers had the opportunity to mine such a wealth of information that promises to provide insights about the complex behaviors of human societies [1,2]. One goal of these researches is to quantitatively uncover the inherent feature of human behaviors and track how our behaviors evolve by mining petabytes of network data.

The studies on human behaviors can be traced since 19th century [3]. The discipline covers psychology, sociology, anthropology, etc., which study different aspects of the nature of human intelligence. However, due to lack of ways of measurement and analysis of large scale of data, the studies mostly focused on individuals or a certain small group or rough qualitative estimation of social behaviors. This was kept almost unchanged until the emergence of modern computing science and network technology. Recent advances in computing and network science have driven the studies on social behaviors to a newly high stage. The universal usage of computational devices leaves huge amount of data that are tightly related with human behaviors. Scientists were able to use computer data, and network data to sense and analyze human behaviors in the society. During this period, a lot of theoretical and practical achievements in social behaviors studies arose. The prevalence of social network in the Internet and the progress in complex network research are best examples for this [4,5]. Meanwhile, with the development and widespread of smartphones and mobile network, they began to show their dominance in sensing human behaviors over the traditional way of mining Internet social media. Therefore, in this paper, we investigate a practical case of mobile network aided social sensing. We study the users’ behaviors with regard to communication, movement, and consumption based on mining a large set of mobile user data. The findings may be significant to instruct marketing strategies for the service providers to increase their revenues and lead them to success through Customer Relationship Management (CRM) [6].

When discussing the communication behaviors, movement behaviors and consumption behaviors of mobile users, we aim at three questions:

1. Does each ARPU (Average Revenue Per User) level imply similar communication behavior?
2. Does each ARPU level imply similar mobility level? And what is the relationship between a user’s consumption capacity and mobility?
3. Do people with similar mobility patterns have similar communication behaviors?

We explore the above three issues in this paper. We notify that the detailed information about human mobility across a large population can be collected by mobile operators that record the closest base station when a call is generated. Herein, we use 16-week-long (from September to December, 2009) calling records of about one million users in a metropolis in China as our data set to conduct research. Each record includes the serving base station’s ID, the start and end times of each phone call (outgoing and incoming calls are distinguished), as well as the monthly billing information of each user. For preserving the users’ privacy, the record of each phone call is anonymous.

We apply both the value-based and behavior-based segmentation methods to investigate the difference between the users who have different or similar ARPU. That is, we firstly divide the users based on their ARPU into different groups as preliminary division. Then, we study their behaviors by using fuzzy c-means (FCM) clustering. The contributions of our paper are:

1. Verify that the same ARPU level users have different behavior patterns, while different ARPU level users may have similar communication behaviors;
2. Prove that the users’ mobility levels have relation with their ARPUs and communication behaviors. People who make less calling or like making calls at night have less mobility than others;
3. Find that the top ARPU level users are the most ‘‘lonely’’ ones, which imply that they are willing to move alone and do not like being with others.

The rest of paper is organized as follows. Section 2 briefly reviews related work. Section 3 introduces a communication behavior indicator and a user segmentation algorithm based on the FCM that are used to study the questions proposed by us. In Section 4, we investigate user behavior patterns and extract salient characteristics that indicate the relationships among users’ consumption capacity, communication time, mobility, and locations. Finally, we further discuss our results and conclude the paper in Section 5.