Inference of social relationship types using co-presence history in a place

Abstract

With the development of mobile devices and wireless network, finding extant social relationships becomes one of the major issues in mobile and social computing environment. In this thesis, motivated by the fact that most of social interactions are made in places which reflect the people`s work and life styles, we take a step forward a novel method of inferring social relationship types such as acquaintance, ordinary relationship, and close relationship by using co-presence history of people in a place. We collected indoor co-location data from 70 participants for a month, setting APs and place servers at various types of places in KAIST campus. To identify time-variant co-present members of a group at a place, we introduce an efficient co-present group monitoring scheme using the group view snapshot (GVS) method. The group view snapshots as meta-information about co-presence of the members are generated by the enter/leave events of members. Based on the data, we designed and explored several indicators useful for inferring social relationship types. Using machine-learning techniques, we found that social relationship between two members can vary according to the places where they made social interactions.