SPECTRAL ANALYSIS OF RICH NETWORK TOPOLOGY IN SOCIAL NETWORKS
1 online resource (164 pages) : PDF
University of North Carolina at Charlotte
Social networks have received much attention these days. Researchers have de- veloped different methods to study the structure and characteristics of the network topology. Our focus is on spectral analysis of the adjacency matrix of the underlying network. Recent work showed good properties in the adjacency spectral space but there are few theoretical and systematical studies to support their findings.In this dissertation, we conduct an in-depth theoretical study to show the close relationship between algebraic spectral properties of the adjacency matrix and vari- ous patterns in a broad range of social networks such as friendship networks, alliance and war networks, and distrusted networks. In our framework, we apply matrix per- turbation theory and approximate the eigenvectors of real graphs by those of the ideal cases. Our theoretical results show that the principal eigenvectors capture the structure of major communities and exhibit them as orthogonal lines/clusters rotated with certain angles from canonical axes. Our results also show that the minor eigen- vectors with skew distributions in values capture weak or subtle signals hidden in local communities. We utilize our theoretical results to develop algorithms for several problems in social network analysis including community partition, anomaly detec- tion and privacy preserving social network reconstruction. Empirical evaluations on various synthetic data and real-world social networks validate our theoretical findings and show the effectiveness of our algorithms.In a nutshell, we theoretically study the patterns in the adjacency spectral space as well as conditions for their existence and explore the application of the spectral properties of the adjacency matrix in different tasks of social network analysis.
MATRIX PERTURBATIONPRIVACY PRESERVING GRAPH RECONSTRUCTIONSIGNAL DETECTIONSOCIAL NETWORKSSPECTRAL ANALYSISUNSIGNED/SIGNED GRAPHS
Lu, AidongRas, ZbyszekWu, WenshengZhang, Zhiyi
Thesis (Ph.D.)--University of North Carolina at Charlotte, 2013.
This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s). For additional information, see http://rightsstatements.org/page/InC/1.0/.
Copyright is held by the author unless otherwise indicated.