Joe’s comment re “role less” fields is the main point.
One other consideration is that the data being summarised can be huge - petabytes of data spread across multiple machines. For this reason we have to work with a sample to avoid blowing memory limits. There’s a danger this sample could be dominated by documents describing comms between a particularly chatty pair to the exclusion of all other pairings. This is why we have the option of a diversity setting - a choice of field where you can limit the number of documents selected for the sample being visualised. For comms data a good choice of field would be a field that contains a combination of source and destination IDs and a size of 1. That means you’d only see one document per unique communicating pair and therefore get better coverage of all the actors. Also remember to increase the number of values returned for each field in the GUI (default is a modest 5, max is 99 per expand step)