In recent years, wireless networks (including WLANs and cellular networks) have been widely deployed in enterprise, public areas and homes. In wireless networks, users are observed to act collaboratively rather than arriving or leaving independently. In this paper, we have analysed real WiFi traces deployed in an enterprise in 2012 and cellular network traces from a urban areas in China in 2013. We find that users in wireless networks are likely to be social related in either time field or spatial field. We also observe that in WLANs, user behavior like leaving together may cause significant AP load unbalance problem, which may result in sub-optimal network throughput and unfair bandwidth allocation among users. Inspired by those observations, we come up with an online heuristic algorithm, which can actively learn users' sociality information and assign them to the optimal APs.