15.08.2015 Views

Introduction to the Modeling and Analysis of Complex Systems

introduction-to-the-modeling-and-analysis-of-complex-systems-sayama-pdf

introduction-to-the-modeling-and-analysis-of-complex-systems-sayama-pdf

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

17.3. CENTRALITIES AND CORENESS 383disseminate a rumor <strong>to</strong> <strong>the</strong> entire network, closeness centrality would probably be a moreappropriate metric <strong>to</strong> use. Or, if you want <strong>to</strong> find <strong>the</strong> most effective person <strong>to</strong> moni<strong>to</strong>r <strong>and</strong>manipulate information flowing within <strong>the</strong> network, betweenness centrality would be moreappropriate, assuming that information travels through <strong>the</strong> shortest paths between people.Eigenvec<strong>to</strong>r centrality <strong>and</strong> PageRank are useful for generating a reasonable ranking<strong>of</strong> nodes in a complex network made <strong>of</strong> directed edges.Exercise 17.5 Visualize <strong>the</strong> Karate Club graph using each <strong>of</strong> <strong>the</strong> above centralitymeasures <strong>to</strong> color <strong>the</strong> nodes, <strong>and</strong> <strong>the</strong>n compare <strong>the</strong> visualizations <strong>to</strong> see howthose centralities are correlated.Exercise 17.6 Generate (1) an Erdős-Rényi r<strong>and</strong>om network, (2) a Watts-Strogatz small-world network, <strong>and</strong> (3) a Barabási-Albert scale-free network <strong>of</strong> comparablesize <strong>and</strong> density, <strong>and</strong> obtain <strong>the</strong> distribution <strong>of</strong> node centralities <strong>of</strong> yourchoice for each network. Then compare those centrality distributions <strong>to</strong> find whichone is most/least heterogeneous.Exercise 17.7 Prove that PageRank with α = 1 is essentially <strong>the</strong> same as <strong>the</strong>degree centrality for undirected networks.A little different approach <strong>to</strong> characterize <strong>the</strong> centrality <strong>of</strong> nodes is <strong>to</strong> calculate <strong>the</strong>ircoreness. This can be achieved by <strong>the</strong> following simple algorithm:1. Let k = 0.2. Repeatedly delete all nodes whose degree is k or less, until no such nodes exist.Those removed nodes are given a coreness k.3. If <strong>the</strong>re are still nodes remaining in <strong>the</strong> network, increase k by 1, <strong>and</strong> <strong>the</strong>n go back<strong>to</strong> <strong>the</strong> previous step.Figure 17.5 shows an example <strong>of</strong> this calculation. At <strong>the</strong> very end <strong>of</strong> <strong>the</strong> process, we seea cluster <strong>of</strong> nodes whose degrees are at least <strong>the</strong> final value <strong>of</strong> k. This cluster is called ak-core, which can be considered <strong>the</strong> central part <strong>of</strong> <strong>the</strong> network.In NetworkX, you can calculate <strong>the</strong> corenesses <strong>of</strong> nodes using <strong>the</strong> core_number function.Also, <strong>the</strong> k-core <strong>of</strong> <strong>the</strong> network can be obtained by using <strong>the</strong> k_core function. Hereis an example:

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!