Research On Analysis Of User Behavior Pattern In University Campus Wireless LANs

With the popularity of smart mobile devices among the students, Campus wireless LANs are widely deployed in the library, laboratory, public academy buildings and student dormitory to provide the students access the internet conveniently. With more and more users connect the network by the wireless LANs, it is more important to understand the user behavior pattern. Firstly, it not only can effectively analyze the status of current network, but also it is significant to network management and optimization in the future. Second, with the rise of location based service [LBS] in the mobile network, analyzing the history user trace in the wireless LAN can classify the user into different social groups in order to provide intelligent and personalized service. Thirdly, analysis of user behavior patterns is helpful to build an available user model and it is also helpful to design the user perceived protocol.In this paper, we study the user behavior pattern in the large-scaled campus wireless LANs. On one hand, we analyze the methods of user behavior patterns research at home and aboard. On the other hand we set up a real measurement platform to monitor current network. In this platform, we adopt SNMP and syslog to get the information from wireless controller and then store the user information into the database. After that, we display some items of user performance in the ways of map and form. Last we collect the syslog information from two AC in our campus, and we analyze some general statistic of user performance:user wireless adapter manufacturer, user online time, user mobility. Furthermore, we mine the social relationship between users. In our social relationship mining research, we apply user profile to represent users in the WLAN trace, and we compute the social similarity based on the encounter-time model. Be different from research before, we do not compute the similarity of all the users. We consider the activeness of users by the AP connection frequency and total online time. We only compute the similarity of high activeness users. After filter and similarity computing, an unsupervised hierarchical clustering algorithm is applied to cluster all users in wireless network based on the similarity matrix. We analyze the distribution of cluster. Last we plot the graph of encounter-relationship about100users and compute the cluster coefficient, disconnected ratio and average path length.