Abstract
Understanding the evolution of relationship among users, through generic interactions, is the key driving force to this study. We model the evolution of friendship in the social network of MobiClique using observations of interactions among users. MobiClique is a mobile ad-hoc network setting where Bluetooth enabled mobile devices communicate directly with each other as they meet opportunistically. We first apply existing topological methods to predict future friendship in MobiClique and then compare the results with the proposed interaction-based method. Our approach combines four types of user activity information to measure the similarity between users at any specific time. We also define the temporal accuracy evaluation metric and show that interaction data with temporal information is a good indicator to predict temporal social ties. The experimental evaluation suggests that the well-known static topological metrics do not perform well in ad-hoc network scenario. The results suggest that to accurately predict evolution of friendship, or topology of the network, it is necessary to utilise some interaction information.