Do communities even exist? Evaluating the robustness of results obtained from a community detection algorithm
Kathleen Gates, University of North Carolina, Quantitative Psychology
This talk describes a method for evaluating the robustness of a final community detection solution. The approach builds from a commonly used heuristic first introduced in physics literature and currently popular in functional MRI applications. The motivation for adapting it for use with sparse, weighted networks came from work on the unsupervised classification of individuals based on their brain processes (as opposed to diagnostic category, age, etc.). Having arrived at a technique that successfully classified individuals into expected communities, there remained a need for an objective evaluation of the solution robustness in an absolute sense. The unsupervised classification technique is briefly introduced with the primary focus of the talk being the evaluation of the resulting community solutions. Examples from functional MRI, daily diary studies, diffusion tensor imaging, and social networks will be presented as well as results from simulated data.
December, 4 2017 | 12:45 pm - 2:00 pm | Gross Hall 230E