For my Fedex Day project, I wrote an application to extract social interaction data from internal Atlassian blog posts and perform several types of Social Network Analysis. Social Network Analysis uses graph theory algorithms to study social relationships among individuals.

I compared the Atlassian blogging community to three other datasets, which can be downloaded for research: interactions among a group of 62 dolphins; social interactions in a Karate club at a US University; and co-appearances among characters in Victor Hugo's Les Miserables. I wrote the program for building the social network graph in Ruby, and I used the R Statistical System and the SNA package to do the social network analysis computations.

The social network graph for blog posts looks like this, where individuals are nodes in the graph and edges are interactions.

I calculated several social network analysis metrics: I calculated "betweenness", which identifies people who bridge various parts of the network; I calculated "closeness", which indicates the strength of someone's "grapevine", for being plugged into information; and I calculated "eigenvector centrality", which is often interpreted to indicated the "power" of a node in the network.

I found distribution of the "betweenness" scores most useful (i.e. discriminating) in comparing the Atlassian community with the other groups. The Atlassian "betweenness" scores were distributed like this:

At the end of the analysis, it was clear that the Atlassian blogging community resembled (you guessed it!) the dolphin community most closely. In the Atlassian and Dolphins graphs, different regions of the graph had many "bridges" between them. In the social network graphs for Les Miserables and the Karate Club, different regions of the graph depended on just a couple of people to "bridge" them. I interpret this as meaning that communication is less "siloed" for the Dolphins and for Atlassian.