Understanding Social Characteristic from Spatial Proximity in Mobile Social Network

Abstract

Over the past decades, cities as gathering places of millions of people rapidly evolved in all aspects of population, society, and environments. As one recent trend, location-based social networking applications on mobile devices are becoming increasingly popular. Such mobile devices also become data repositories of massive human activities. Compared with sensing applications in traditional sensor network, Social sensing application in mobile social network, as in which all individuals are regarded as numerous sensors, would result in the fusion of mobile, social and sensor data. In particular, it has been observed that the fusion of these data can be a very powerful tool for series mining purposes. A clear knowledge about the interaction between individual mobility and social networks is essential for improving the existing individual activity model in this paper. We first propose a new measurement called geographic community for clustering spatial proximity in mobile social networks. A novel approach for detecting these geographic communities in mobile social networks has been proposed. Through developing a spatial proximity matrix, an improved symmetric nonnegative matrix factorization method (SNMF) is used to detect geographic communities in mobile social networks. By a real dataset containing thousands of mobile phone users in a provincial capital of China, the correlation between geographic community and common social properties of users have been tested. While exploring shared individual movement patterns, we propose a hybrid approach that utilizes spatial proximity and social proximity of individuals for mining network structure in mobile social networks. Several experimental results have been shown to verify the feasibility of this proposed hybrid approach based on the MIT dataset.