Pranav Pamidighantam's Dissertation Defense

As social networks and social media continue to gain popularity, understanding how information spreads across these networks grows in importance. A critical component of understanding information spread is determining the important actors; those who are apt at spreading information and those who readily accept it. Three primary computational challenges exist when designing methodologies to identify these actors: large-scale networks need to be efficiently processed and analyzed, rich information about the attributes of actors and their relationships should be able to be utilized, and dynamic networks where the relationships between actors evolve have to be considered. To address these challenges, we propose the Framework for Influence and Susceptibility Computation, a computational framework for the categorization, design, and analysis of methodologies to find important actors in information diffusion scenarios. We hypothesize the framework to have four capabilities: 1) categorization of existing methodologies to describe their components, 2) design of novel methodologies, 3) analysis of efficacy of design choices, and 4) identification of potential strengths and weaknesses of methodologies in specific problem contexts. To test these capabilities, four sets of methodologies are created using the framework and experimentally tested over a range of information spreading contexts. We develop a methodology for predicting the susceptibility of actors in large-scale and attributed networks and extend this methodology for application to dynamic networks. In addition, we design a methodology to find opinion leaders using various network structures. Finally, we use the aforementioned susceptibility predictions to create a set of methodologies to find influential actors across a range of networks. Throughout this dissertation, we test the efficacy of various design choices within these methodologies, and develop an understanding of how network structure can affect the performance of methodologies to find important actors. We demonstrate the utility of the proposed computational framework and move towards a deeper understanding of how information spreads.

Questions? Contact Pranav Pamidighantam. 

In reference to IEEE copyrighted material which is used with permission in this thesis, the IEEE does not endorse any of University of Illinois at Urbana-Champaign’s products or services. Internal or personal use of this material is permitted. If interested in reprinting/republishing IEEE copyrighted material for advertising or promotional puposes or for creating new collective works for resale or redistribution, please obtain a License from RightsLink. If applicable, University Microfilms and/or ProQuest Library, or the Archives of Canada may supply single copies of the dissertation.