Diffusion, Infection and Social (Information) Network Database
Diffusion, Infection and Social (Information) Network Database
Files
Publication or External Link
Date
2015
Authors
Kang, Chanhyun
Advisor
Subrahmanian, V.S.
Citation
DRUM DOI
Abstract
Research to analyze diffusive phenomena over large rich datasets has received considerable attention in recent years. Moreover, with the appearance and proliferation of online social network services, social (information) network analysis and mining techniques have become closely intertwined with the analysis of diffusive and infection phenomena. In this dissertation, we suggest various analysis and mining techniques to solve problems related to diffusive and infection phenomena over social (information) networks built from various datasets in diverse areas. This research makes five contributions. The first contribution is about influence analysis in social networks for which we suggest two new centrality measures, Diffusion Centrality and Covertness Centrality. Diffusion Centrality quantifies the influence of vertices in social networks with respect to a given diffusion model which explains how a diffusive property is spreading. Covertness Centrality quantifies how well a vertex can communicate (diffuse information) with (to) others and hide in networks as a common vertex w.r.t. a set of centrality measures. The second contribution is about network simplification problems to scale up analysis techniques for very large networks. For this topic, two techniques, CoarseNet and Coarsened Back and Forth (CBAF), are suggested in order to find a succinct representation of networks while preserving key characteristics for diffusion processes on that network. The third contribution is about social network databases. We propose a new network model, STUN (Spatio-Temporal Uncertain Networks), whose edges are characterized with uncertainty, space, and time, and develop a graph index structure to retrieve graph patterns over the network efficiently. The fourth contribution develops epidemic models and ensembles to predict the number of malware infections in countries using past detection history. In our fifth contribution, we also develop methods to predict financial crises of countries using financial connectedness among countries.