An indirect approach for blockmodeling temporal networks
Vladimir BatageljWe first discuss different aspects of the temporal network blockmodeling. In the talk, we propose an approach to a simple case where the clusters (roles) do not change through time. In this case, we can obtain a clustering using an adapted hierarchical clustering method, similarly as in the case of clustering with relational constraint (Ferligoj and Batagelj, 1983; Doreian et al, 2020, p. 79-81, 90-95). Three auxiliary questions should be answered:
- how to measure dissimilarity between units/clusters;
- definition of the fusibility predicate that determines which units/clusters can be joined;
- how to update the links between the new (merged) node and remaining nodes.
We give some answers to these questions. The right number of clusters can be selected on the basis of the obtained dendrogram. To get the blockmodel we need to determine also temporal descriptions of model links.
We will illustrate the proposed approach by applying it to some temporal networks such as Franzosi's violence network and the Terror news network.