Abstract
This thesis has been a joint project between Universidad Carlos III de Madrid and Telefonica Research (Spain). Specifically, the research has been conducted at the GISC (Grupo Interdisciplinar de Sistemas Complejos), group of Universidad Carlos III and at the analytics and data mining and user modelling research teams of Telefonica Research. The main interest of this research has been in understanding and characterizing large networks of human interactions as continuously changing objects, which members appear and disappear over time and which interactions are characterized by temporal correlations and inhomogeneities. This constitutes a very challenging and novel topic. In fact, although many real social networks are temporal or dynamical networks, which elements and properties continuously change over time, traditional approaches to social network analysis are essentially static: ties (and tie weights) are given by the aggregated activity observed in a given time period, nodes and ties are considered persistent over time, temporal inhomogeneities and correlations between interaction events are neglected, etc. Within this frame, therefore, the time dimension of human behavior has typically been projected out. Although much effort has been devoted in the last years to characterize the temporal patterns of human interaction, a general understanding of how dynamically model real social networks is still missing. In this thesis we contribute to advancing the state of the art in this area by investigating the instantaneous, instead than the aggregated, contact network and by analyzing the role of temporal activity patterns of human interaction in the description and modeling of real social networks. Specifically, we investigated the role that topological and, in particular, temporal patterns of human interaction play in three main topics of social network analysis and data mining: the characterization of time (or attention) allocation in social networks, the prediction of link decay and/or persistence and the analysis and modeling of information spreading phenomena. To this end, we have analyzed large anonymized data sets of phone call communication traces (Call Detail Records or CDR) over long periods of time. Access to these observations was granted by Telefonica Research. The availability of empirical data about such massive networks allowed us to analyze and measure global features of human behavior and interaction and to characterize phenomena and tendencies that might be invisible at small scale. At the same time, the fine-grained resolution of the datasets we had access to and the fact that they cover a large sample of the population, ensure the significance and universality of our findings. The findings that emerge from our research indicate that the observed inhomogeneities and correlations of human temporal patterns of interactions significantly affect the current view of social networks, shifting from a very steady to a highly complex entity. Temporal patterns of communication are essential not only for a better characterization of the inherent properties of human behavior, but also, and more importantly, for the understanding and modeling of all those phenomena which are triggered by the way in which people communicate and behave. Examples are diffusion of epidemics, information spreading, opinion and influence phenomena and group formation. Our results indicate the necessity to incorporate temporal patterns of communication in the analysis of social networks: since structure and dynamics are tightly coupled, the analysis and modeling of human behavior has to factor in both. The work of this thesis combines data mining, the analysis of large datasets, theoretical modeling, simulations and experiments on empirical data. In addition, this also has a wide range of applications in many business sectors. In particular, at Telefonica Research, part of our techniques and findings have been successfully applied to areas such as social networks analysis, modeling human influence, customer segmentation and targeting in viral marketing campaigns. We believe this work has made a contribution to understanding and modeling real social networks and and we are confident that it will encourages further research in this field. --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.