Social Interaction Discovery: A Simulated Multiagent Approach

Abstract

Abstract. Social interaction inference is a problem that has been of interest in the past few years. The intrinsic mobility patterns followed by humans present a number of challenges that range from interaction inference to identification of social relationships linking individuals. An intuitive approach is to focus on the similarity of mobility patterns as an indicator of possible social interaction among individuals. By recording the access points observed at each unit of time along with the strength of the signals received, individuals may be group based on similar walking patterns shared on space and time. In this paper, an implementation of a multiagent simulation of a University-like environment is tested using NetLogo and a methodology that consists of two phases: 1) Cluster Analysis and 2) Construction of Social Networks is used to discover possible interactions among individuals. The first phase consists of a number of clustering methods that are used to identify individuals that are more closely related given the characteristics that describe their mobility patterns obtained from simulated Wifi data. In the second phase, users belonging to the same cluster are linked within a social network, meaning that there is possible ongoing social interaction or tie that might link the individuals.