Clustering Methods for Network Adjacency Data

Clustering analysis aims to detect the topological community-structure of networks (connected graphs with n vertices and m edges), and studies inherent relations behind partitions. In this thesis, we consider far-reaching model-free clustering algorithms including Girvan and Newman’s edge betweenness algorithm, Zhou’s dissimilarity algorithm, the Walktrap algorithm, the leading eigenvector algorithm, the fast-greedy algorithm and the Louvain method, in relation to each other. We also introduce a unified and natural approach, based on a newly defined dissimilarity, to clustering subsets of vertices, which are considered to be active (occupied, selected) within a network. The informativeness and effectiveness of these algorithms are considered in relation to well-known real-world test-case datasets (with reasonable ground truths) including Zachary’s karate network, the U.S. football network, a dolphins social network, a macaque brain network, a cat cortex network, and a political books network.

Grantor dc:publisher

Wake Forest University

Year dc:date.issued

2016

Author dc:creator

• Zhang, Teng

• Clustering algorithms

Language dc:language.iso

en

Handle dc:identifier.uri

http://hdl.handle.net/10339/62641

OAI identifier oai:identifier

oai:wakespace.lib.wfu.edu:10339/62641