Advances in Network Clustering and Blockmodeling

This chapter unfolds different aims underpinning community detection – in a relaxed form that includes assortative as well as disassortative group structures with dense and sparse internal connections, respectively – and discusses how the resulting problem perspectives relate to various applications. It focuses on four broad perspectives that have served as motivation for community detection in the literature: the cut-based perspective that minimizes a constraint such as the number of links between groups of nodes, the clustering perspective that maximizes internal density in groups of nodes, the stochastic block model perspective that identifies groups of nodes in which nodes are stochastically equivalent, and the dynamical perspective that identifies groups of nodes in which flows stay for a relatively long time such that they form building blocks of dynamics on networks. While this categorization is not unique, the authors believe that it can help clarify concepts about community detection.