Title :
A Spectrum-Based Framework for Quantifying Randomness of Social Networks
Author :
Ying, Xiaowei ; Wu, Leting ; Wu, Xintao
Author_Institution :
Dept. of Software & Inf. Syst., Univ. of North Carolina at Charlotte, Charlotte, NC, USA
Abstract :
Social networks tend to contain some amount of randomness and some amount of nonrandomness. The amount of randomness versus nonrandomness affects the properties of a social network. In this paper, we theoretically analyze graph randomness and present a framework which provides a series of nonrandomness measures at levels of edge, node, subgraph, and the overall graph. We show that graph nonrandomness can be obtained mathematically from the spectra of the adjacency matrix of the network. We derive the upper bound and lower bound of nonrandomness value of the overall graph. We investigate whether other graph spectra (such as Laplacian and normal spectra) could also be used to derive a nonrandomness framework. Our theoretical results showed that they are unlikely, if not impossible, to have a consistent framework to evaluate randomness. We also compare our proposed nonrandomness measures with some traditional measures such as modularity. Our theoretical and empirical studies show our proposed nonrandomness measures can characterize and capture graph randomness.
Keywords :
graph theory; matrix algebra; social networking (online); Laplacian spectra; adjacency matrix; graph nonrandomness; graph randomness; graph spectra; nonrandomness measures; normal spectra; overall graph; randomness quantification; social networks; spectrum-based framework; Eigenvalues and eigenfunctions; Quantization; Random processes; Social network services; Randomness measures; graph spectra; social networks.;
Journal_Title :
Knowledge and Data Engineering, IEEE Transactions on
DOI :
10.1109/TKDE.2010.218