2015
Ce document est lié à :
info:eu-repo/grantAgreement/EC/FP7/313847/EU/Globalization, regionalization, urbanization: an analysis of the worldwide maritime network since the early 18th century/WORLD SEASTEMS
info:eu-repo/semantics/OpenAccess
Charles Bouveyron et al., « Cluster identification in maritime flows with stochastic methods », HAL-SHS : géographie, ID : 10670/1.5vdgk5
We consider two statistical models for network analysis: SBM and RSM. The stochastic block model (SBM) is one of the most popular models used to look for clusters of nodes in networks. Its flexibility allows it to model a wide variety of networks (communities, stars, etc.) and to summarise the network with a few interpretable parameters. The random subgraph model (RSM) is a more recent approach which extends SBM to networks with categorical edges and for which a partition into subgraphs is known (a geographical partition for instance). Thus, RSM allows to find clusters which correspond to complex latent groups and not to obvious (geographical, for instance) clusters. We finally present the application of both SBM and RSM to a maritime flow network extracted from the Lloyd’s List. The network analysis is done using the mixer and Rambo packages for the R software and some code listings illustrate the practical use of their main functions.