Seminario del prof. Ales Ziberna, Università di Ljubljana, visiting professor presso il DiSPeS.

Abstract: The talk will start with the introduction to blockmodeling (a method for partitioning networks). Emphasis will be on generalized blockmodeling including blockmodeling of valued networks, although connections with community detection, stochastic blockmodeling or block mixture models will also be discussed. In the main part of the talk, blockmodeling of multilevel network data will be presented. 

Multilevel network data consist of networks that are measured on at least two levels (e.g. between organizations and people) and information on ties between these levels (e.g. information on which people are members of which organizations). Several approaches are

possible: a) separate analysis of the levels; b) transforming all networks to one level and blockmodeling on this level using information from both/all levels; c) truly multilevel approach, where both/all levels and ties between them are modeled at the same time. 

Here we will focus on the last approach. The approach was initially developed within generalized blockmodeling. However, for larger networks (several hundred units), this approach is not appropriate. 

Therefore, algorithms for multilevel blockmodeling based on two-mode k-means and stochastic blockmodeling will also be discussed.

In addition (time permitting), use of the same procedure for blockmodeling networks in several time points will be presented. This procedure is especially useful for blockmodeling networks in a few time points where we want the groups from different time points to be connected, while not necessary the same, especially when not all units are present in all time points.