Networkx Finding communities of directed graph

Question

I am currently graphing and visualizing a directional social network. There is a statistic (modularity) in an open source visualization tool called Gephi ( http://gephi.github.io/ ) that allows me to partition nodes by separating them into communities.

I'd like to accomplish this task (finding communities) with my graph before I visualize it with Gephi so that I can run various statistics on each community that the algorithm finds.

However, each community algorithm that I have found does not operate on directed graphs. I found two different algorithms to work with that won't work with diGraphs:

1. Aynaud's community algorithm based on dendograms at http://perso.crans.org/aynaud/communities/api.html
2. NetworkX's own find k cliques at https://networkx.github.io/documentation/latest/reference/generated/networkx.algorithms.community.kclique.k_clique_communities.html

This raises several questions, namely:

1. Is this a limitation of the type of community finding algorithm that I want to use?
2. If so, is Gephi using its algorithm by removing the directionality of the graph before looking for communities?
3. Are there other algorithms I should be looking for?
4. Are there other libraries where this is already solved?

Thank you!

Answer 1

There are several algorithms in Machine Learning that classify nodes into groups. Usually there is some data or node name that makes a node distinctive as part of one group or another. There might be other ways to break up the nodes besides data information like the number of links (which is a type of calculated data used in the pagerank algorithm). There are several to try based on the criteria like Support Vector machines or ensemble methods like random forests. There is also ways to reduce the dimentionality of your data using Principle Component Analysis. There is some excellent documentation on the scikit-learn page. http://scikit-learn.org/stable/documentation.html