As a side note, the dataset is actually quite dirty: there are a lot of duplicated names and locations. Many names are listed both with and without honorifics. Many officers are listed both by first name/last name and last name/first name (and some full names also include middle names). There are numerous typos in names.

To answer Moses' question, I would suggest treating the whole network as tripartite (officer+intermediary / address / entity) and then project it to get either a network of entities or a network of officers/intermediaries.

On Sat, May 14, 2016 at 9:55 AM, Moses Boudourides <[log in to unmask]> wrote:

I would appreciate if someone is willing to share any ideas about a
meaningful aggregation scheme for relations. Of course, one could
disregard any sort of relational aggregation and treat the network as
a multilayered (multiplex) one, although the size and the complexity
of the Panama Papers network appear to be rather restraining.

Dmitry Zinoviev
Professor of Computer Science
Suffolk University, Boston, MA 02114