A Study of Familiar Strangers in Temporal Interaction Networks

Abstract

There are rich and varied social ties between people in our daily lives. The familiar stranger is one of those social ties, and has attracted a lot of attention since the concept was proposed. Using knowledge of network science to analyze familiar strangers has broad prospects. The thesis mainly focus on the familiar stranger classifier and the application of the characteristics of familiar strangers. Detailed studies contain three aspects: the influence of different social ties on the epidemic spreading process in the multiplex network, large-scale temporal interaction network construction and designing more widely applicable familiar stranger classifier. The main contents are organized as follows: In recent years, many studies have focused on the interrelation between the epidemic spreading and the spread of awareness to prevent infections. A multiplex network is utilized to depict this scenario. The upper layer of the multiplex network represents an online social network, where the awareness of epidemic spreads. The lower layer of the multiplex network represents an off-line social network, where the epidemic spreads among individuals. However, most previous studies neglected that the online social network has diverse social ties, such as the dense social ties (in-roles and friends) and the sparse social ties (familiar strangers and strangers). In Chapter 3, we study how the online social ties impact on epidemic spreading in the off-line social network. We assume that besides the same dense in-roles and friends, an online social network could have only familiar strangers, only strangers, or both. We analyze the influence of the effective epidemic spreading rate and the awareness spreading rate on the infection fraction at the steady state with three empirical datasets. An interesting finding is that an online social network containing strangers can reduce the epidemic prevalence more than that containing familiar strangers. With the advancement of data collection, storage, and analysis technologies, more and more researchers have turned to the study of large-scale networks. Some studies in sociology show that familiar strangers arise from the periodicity of individual’s behavior and widely exist in our daily lives. In order to use the knowledge of the network science to study the familiar stranger, we need first construct a large-scale interaction network. In the Chapter 4, we use the Shanghai public transportation card dataset and subway operation dataset to build a subway passenger interaction network. We find that the behavior of passengers is periodic and has certain regularity, but it is more unstable than the behavior of students on the campus. Existing algorithm has more or less defects, such as requiring predetermined parameters, high calculation complexity, lack of validation, etc. In response to the problems, we propose a familiar stranger identifying algorithm based on the edit distance (FSCED) in Chapter 5. The correctness of the classification results of our algorithm is validated on 3 campus datasets and the subway passenger interaction network. In general, we found the role of familiar strangers in the epidemic spreading process. Online social networks containing strangers are more able to inhibit the spread of viruses in offline networks than those containing familiar strangers. At the same time, we improved the existing familiar stranger detection algorithm. The FSCED algorithm we proposed can simultaneously dig out interaction patterns in campus scenes with individual behavior patterns and commuting scenes with complex individual behavior patterns, and has lower computational complexity. At the same time, the FSCED algorithm we proposed can dig out interaction regularities in both campus scenes with fixed individual behaviors and commuting scenes with complex individual behaviors. The computational complexity of the FSCED is lower than existing algorithms.