1325. sredin seminar: Fabio Ashtar Telarico: Approaches to dynamic (temporal) SBM: The case of Slovene co-authorship networks (1991-2020)

Approaches to dynamic (temporal) SBM: The case of Slovene co-authorship networks (1991-2020)

Recently, the research on blockmodeling has seen the development on a sprawling literature on networks representing relations between units in two or more time periods (so-called dynamic or temporal networks). Crucially, the literature offers several techniques to blockmodel such networks which vary in important ways (e.g., the definition of the network, mathematical structure); the efforts to compare them in empirical application has been limited.

Using data from the COBISS database, this presentation offers an overview of the results produced on the same network by three approaches to stochastic blockmodeling (SBM) selected on the basis of research by Prof. Žiberna and Dr. Cugmas. At first, particular attention is given to the procedure for selection and preparation of the underlying data. Subsequently, a discussion of the chosen approaches and the resulting blockmodels is offered. First, the SBM for linked networks as operated by the R package StochBlocTest, Second, the implementation of that the SBM for generalised multipartite networks offered in the R package GREMLINS. Finally, one of the approaches designed ad-hoc for dynamic networks, the one implemented in the package dynsbm, is also considered. Notably, the first two approaches imply a conceptualisation of the network (as linked or generalised-multipartite, respectively) that makes them suitable for other types of networks, too. Whereas the third one is only suitable for dynamic networks.

In conclusion, preliminary results suggest that the approaches for linked and multiplartite networks can yield quite similar results. Moreover, both are quite more flexible than the SBM for dynamic networks due to the latter's assumption that the number of clusters is constant over time.