Graph Theory and Complex Network Analysis Group, Technical University Kaiserslautern
Network Analysis Literacy
Network analysis provides the uniform framework to analyze big data sets from disciplines as far apart as medicine, sociology, human complex problem solving, cancer biology, and archaeology. Once the problem is represented as a network, various centrality measures, clustering algorithms, and statistical models can be applied to find the most central nodes, the densest subgraphs, or the strongest motifs.
Most of us have heard sentences like this for so long, that we do not even question them anymore. In this talk I will show that while all methods are in principle applicable, they come with their own modeling assumptions that might yield their results useless for a specific network. I will prove the point by first showing that airports with a low degree and high betweenness centrality are not necessarily anomalous in an economic sense before I solve the riddle why market basket analysis assumes that Pretty Woman is a good recommendation for somebody who loves Star Wars V. Understanding when to use which method is what I call "network analysis literacy". To make us all more literate, I finally advocate for a more transparent desciption of network analytic methods and their implicit modeling assumptions to enable an informed choice of the best measure for a given network analytic question.