Graph clustering in industrial networks

Abstract

The present work investigates clustering of a graph-based representation of industrial connections derived from international trade data by Hidalgo et al (2007) and confirms existence of around ten industrial clusters that are reasonably consistent with expected historical patterns of diffusion of innovation and technology. This supports the notion that technological development occurs in sequential innovation waves. The clustering method developed in this work follows conceptual ideas of Lambiotte and Barahona (2009), who suggested to use random walk to assess a hierarchical structure of network communities where different levels of the hierarchy correspond to different diffusion times. We, however, implement these ideas differently to match physics of the problem under consideration and introduce a hierarchal clustering procedure that is combined with convenient resorting of the elements. An equivalent spectral interpretation of the clustering is also given and discussed in the paper.