In the context of network science community detection is an important aspect of analysis.
Most of the approaches are iterative: they scan the network to see which nodes tend to be more densely connected together than with the nodes from the rest of the network. These nodes are then organized into communities.
For example, InfraNodus uses Louvain community detection algorithm (Blondel et al 2008). It detects the nodes (words) that co-occur more often together than others and puts them into the same community (designated with a distinct color each). The communities are then ranged by size.
We then range those nodes either according to the degree or the number of connections that each node has (default measure — correlates with the node's frequency) or betweenness centrality.