UMBC ebiquity
Twitter Social Network Analysis

Twitter Social Network Analysis

Akshay Java, 1:00pm 19 April 2007

In the recent series of posts, we have presented Twitter Goolgle Maps mashup, a Twitter search and buzz tracking tool called Twitterment and analysis of geolocation information from the twitter dataset. By providing a neat API, Twitter has enabled researchers to get a better understanding of Microblogging.

In this post, I have used the Large Graph Layout (LGL) tool to visualize the social network on Twitter. Following is a graph constructed using contacts from about 25K users. Notice that there is a link connecting two users if either one has the other as a friend and hence it is an undirected graph (of about 250K edges).

Compare this to the following graph that is constructed using only users who are mutually acquainted. i.e. A knows B and also B knows A.

I find that visualizing such large graphs is quite a challenge and to glean meaningful information from it is even more difficult. However there are a few insights one can gain from this:

  • Interestingly, there are a number of users who are trying to win a popularity contest of some sorts! The complete list of users ranked by the number of friends they have is shown here.
  • A number of bloggers and (perhaps fake?) celebrity profiles have a huge fan following in Twitter. Here is a list of users ranked by number of followers.
  • The two graphs shown above look very different on account of the fact that users with public profiles get a lot of followers whom they might not really know and would hence never add them as an acquaintance (well, in most cases atleast). But to really understand what the differences are one would need to look at the community structure and properties of the two graphs.

Finally, for completeness, here is a list of users ranked according to their PageRank scores. It is noticeably similar to the rankings generated by Twitterholic. This can be explained by the fact that local metrics (like number of followers) in a social network are a good first order approximation of rank. Dr. Finin made me aware of research by social network expert Valdis Krebs, who uses “reach” as a measure in human social networks. Here a person’s reach is the number of other people that are within N links in the network where N is usually 1, 2 or 3 for human networks. So, Twitterholic rank for example is the case with N equal to 1.
[Thanks Eytan and Matt for suggestions on Graph Visualization tools. Related: Matt, Bruno’s posts on network visualization of Belgian bloggers]

Comments are closed.