In this thesis we address the close interplay among mobility, offline relationships and online interactions and the related human networks at different dimensional scales and temporal granularities. By generally adopting a data-driven approach, we move from small datasets about physical interactions mediated by human-carried devices, describing small social realities, to large-scale graphs that evolve over time, as well as from human mobility trajectories to face-to-face contacts occurring in different geographical contexts. We explore in depth the relation between human mobility and the social structure induced by the overlapping of different people's trajectories on GPS traces collected in urban and metropolitan areas. We define the notions of geo-location and geo-community which are operational in describing in a unique framework both spatial and social aspects of human behavior. Through the concept of geo-community we model the human mobility adopting a bipartite graph. Thanks to this graph representation we can generate a social structure that is plausible w.r.t. the real interactions. In general the modeling approach have the merit for reporting the mobility in a graph-theoretic framework making the study of the interplay mobility/sociality more affordable and intuitive. Our modeling approach also results in a mobility model, Geo-CoMM, which lies on and exploits the idea of geo-community. The model represents a particular instance of a general framework we provide. A framework where the social structure behind the preferred-location based mobility models emerges. We validate Geo-CoMM on spatial, temporal, pairwise connectivity and social features showing that it reproduces the main statistical properties observed in real traces. As concerns the offline/online interplay we provide a complete overview of the close connection between online and offline sociality. To reach our goal we gather data about offline contacts and social interactions on Facebook of a group of students and we propose a multidimensional network analysis which allows us to deeply understand how the characteristics of users in the distinct networks impact each other. Results show how offline and Facebook friends are different. This way we confirm and worsen the general intuition that online social networks have shifted away from their original goal to mirror the offline sociality of individuals. As for the role and the social importance, it becomes apparent that social features such as user popularity or community structure do not transfer along social dimensions, as confirmed by our correlation analysis of the network layers and by the comparison among the communities. In the last chapters we analyze the evolution of the online social network from a physical time perspective, i.e. considering the graph evolution as a graph time-series and not as a function of the network basic properties (number of nodes or links). As for the physical time in a user-centric viewpoint, we investigate the bursty nature of the link creation process in online social network. We prove not only that it is a highly inhomogeneous process, but also identify patterns of burstiness common to all nodes. Then we focus on the dynamic formation of two fundamental network building components: dyads and triads. We propose two new metrics to aid the temporal analysis on physical time: link creation delay and triangle closure delay. These two metrics enable us to study the dynamic creation of dyads and triads, and to highlight network behavior that would otherwise remain hidden. In our analysis, we find that link delays are generally very low in absolute time and are largely independent of the dates people join the network. To highlight the social nature of this metric, we introduce the term \textit{peerness} to quantify how well linked users overlap in lifetimes. As for triadic closure delay we first introduce an algorithm to extract of temporal triangle which enables us to monitor the triangle formation process, and to detect sudden changes in the triangle formation behavior, possibly related to external events. In particular, we show that the introduction of new service functionalities had a disruptive impact on the triangle creation process in the network.
FROM SMALL-WORLDS TO BIG DATA:TEMPORAL AND MULTIDIMENSIONAL ASPECTS OF HUMAN NETWORKS / M. Zignani ; tutor: S. T. Gaito ; coordinatore: E. Damiani. DIPARTIMENTO DI INFORMATICA, 2014 Mar 18. 26. ciclo, Anno Accademico 2013. [10.13130/zignani-matteo_phd2014-03-18].
FROM SMALL-WORLDS TO BIG DATA:TEMPORAL AND MULTIDIMENSIONAL ASPECTS OF HUMAN NETWORKS
M. Zignani
2014
Abstract
In this thesis we address the close interplay among mobility, offline relationships and online interactions and the related human networks at different dimensional scales and temporal granularities. By generally adopting a data-driven approach, we move from small datasets about physical interactions mediated by human-carried devices, describing small social realities, to large-scale graphs that evolve over time, as well as from human mobility trajectories to face-to-face contacts occurring in different geographical contexts. We explore in depth the relation between human mobility and the social structure induced by the overlapping of different people's trajectories on GPS traces collected in urban and metropolitan areas. We define the notions of geo-location and geo-community which are operational in describing in a unique framework both spatial and social aspects of human behavior. Through the concept of geo-community we model the human mobility adopting a bipartite graph. Thanks to this graph representation we can generate a social structure that is plausible w.r.t. the real interactions. In general the modeling approach have the merit for reporting the mobility in a graph-theoretic framework making the study of the interplay mobility/sociality more affordable and intuitive. Our modeling approach also results in a mobility model, Geo-CoMM, which lies on and exploits the idea of geo-community. The model represents a particular instance of a general framework we provide. A framework where the social structure behind the preferred-location based mobility models emerges. We validate Geo-CoMM on spatial, temporal, pairwise connectivity and social features showing that it reproduces the main statistical properties observed in real traces. As concerns the offline/online interplay we provide a complete overview of the close connection between online and offline sociality. To reach our goal we gather data about offline contacts and social interactions on Facebook of a group of students and we propose a multidimensional network analysis which allows us to deeply understand how the characteristics of users in the distinct networks impact each other. Results show how offline and Facebook friends are different. This way we confirm and worsen the general intuition that online social networks have shifted away from their original goal to mirror the offline sociality of individuals. As for the role and the social importance, it becomes apparent that social features such as user popularity or community structure do not transfer along social dimensions, as confirmed by our correlation analysis of the network layers and by the comparison among the communities. In the last chapters we analyze the evolution of the online social network from a physical time perspective, i.e. considering the graph evolution as a graph time-series and not as a function of the network basic properties (number of nodes or links). As for the physical time in a user-centric viewpoint, we investigate the bursty nature of the link creation process in online social network. We prove not only that it is a highly inhomogeneous process, but also identify patterns of burstiness common to all nodes. Then we focus on the dynamic formation of two fundamental network building components: dyads and triads. We propose two new metrics to aid the temporal analysis on physical time: link creation delay and triangle closure delay. These two metrics enable us to study the dynamic creation of dyads and triads, and to highlight network behavior that would otherwise remain hidden. In our analysis, we find that link delays are generally very low in absolute time and are largely independent of the dates people join the network. To highlight the social nature of this metric, we introduce the term \textit{peerness} to quantify how well linked users overlap in lifetimes. As for triadic closure delay we first introduce an algorithm to extract of temporal triangle which enables us to monitor the triangle formation process, and to detect sudden changes in the triangle formation behavior, possibly related to external events. In particular, we show that the introduction of new service functionalities had a disruptive impact on the triangle creation process in the network.File | Dimensione | Formato | |
---|---|---|---|
phd_unimi_R09035.pdf
accesso aperto
Tipologia:
Tesi di dottorato completa
Dimensione
14.71 MB
Formato
Adobe PDF
|
14.71 MB | Adobe PDF | Visualizza/Apri |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.