In order to align nodes on a graph, we should use a metric that would enable us to get a meaningful insight about the relations between these nodes.
This iterative algorithm works in the following way:
1. Identify the nodes with the highest number of connections (the hubs) and push them apart from each other;
2. Pull the nodes that are connected to those hubs towards each other;
As a result, after multiple iterations, the most influential nodes are spread in the representation space while the smaller nodes that are connected to them are pulled towards those hubs, forming communities.
This layout is correlated to the communities detected using the method proposed by Blondel et al (which are shown with distinct colors on the graph), so most likely the topics identified with InfraNodus will also be spatially located on the graph in a specific cluster.