Applications of Cohesive Subgraph Detection Algorithms to Analyzing Socio-Technical Networks

Abstract

Socio-technical networks can be productively modeled at several granularities, including the interaction of actors, how this interaction is mediated by digital artifacts, and sociograms that model direct ties between the actors themselves. Cohesive subgraph detection algorithms (CSDA, a.k.a. “community detection algorithms”) are often applied to sociograms, but also have utility in analyzing graphs corresponding to other levels of modeling. This paper illustrates applications of CSDA to graphs modeling interaction and mediated association. It reviews some leading candidate algorithms (particularly InfoMap, link communities, the Louvain method, and weakly connected components, all of which are available in R), and evaluates them with respect to how useful they have been in analyzing a large dataset derived from a network of educators known as Tapped In. This practitioner-oriented evaluation is a complement to more formal benchmark based studies common in the literature.