Statistical Modelling 22 (4) (2022), 273–296

Block models for generalized multipartite networks: Applications in ecology and ethnobiology

Avner Bar-Hen,
CNAM,
75003 Paris,
France.
e-mail: avner@cnam.fr

Pierre Barbillon,
Université Paris-Saclay,
AgroParisTech,
INRAE,
UMR MIA-Paris,
75005, Paris,
France.

Sophie Donnet
Université Paris-Saclay,
AgroParisTech,
INRAE,
UMR MIA-Paris,
75005, Paris,
France.



Abstract:

Generalized multipartite networks consist in the joint observation of several networks implying some common pre-specified groups of individuals. Such complex networks arise commonly in social sciences, biology, ecology, etc. We propose a flexible probabilistic model named Multipartite Block Model (MBM) able to unravel the topology of multipartite networks by identifying clusters (blocks) of nodes sharing the same patterns of connectivity across the collection of networks they are involved in. The model parameters are estimated through a variational version of the Expectation–Maximization algorithm. The numbers of blocks are chosen using an Integrated Completed Likelihood criterion specifically designed for our model. A simulation study illustrates the robustness of the inference strategy. Finally, two datasets respectively issued from ecology and ethnobiology are analyzed with the MBM in order to illustrate its flexibility and its relevance for the analysis of real datasets.

Keywords:

networks, Latent block models, stochastic block models, variational EM, model selection, ecology

Downloads:

The code to reproduce the results is available on github.

back