Visualising a monetised Twitter network

This is just a little experiment with nodeXL, inspired by this example of using it to visualise a Twitter network. NodeXL is a very nice social network analysis (SNA) and visualisation tool. It works from Microsoft Excel, and is very light and easy to use. The NodeXL Tutorial provides instructions on how to use it.

One thing that's particularly nice, for an SNA neophyte like myself, is that nodeXL can both search the net and do the visualisation (you can do this on VOSON too, though). And you can search Twitter too.

Many people on the Malaysian twitterverse will have noticed #xpaxblackberry coming up fairly often recently, and it seems clear that Xpax had purchased the help, perhaps via ChurpChurp, of various key bloggers/tweeters to get the word out. In addition, Xpax was organising an event last Saturday (which I was able to go to, after entering a competition with Nuffnang) to launch their new prepaid Blackberry service.

So - I decided to see what would happen if I put the search term - "xpaxblackberry" into NodeXL.

This is what I got on the 8th October - two days before the launch party


This represents the tweeters who mentioned 'xpaxblackberry' in their tweet, and the lines represent who follows whom, within that group.

The size of the picture is relative to the "Betweenness centrality" of the tweeter: i.e. some people are more connected to other people, either directly or via other people, so they are 'in between' more people. For example: if I know Joe, Peter, and Jane, but none of them know each other, then I have a greater 'betweenness' value.

So, in the above graph, we can see that the four tweeters with the greatest centrality are @kennysia (BC value = 1), @benjern (BC value = 0.876), @julesisapen (BC value = 0.703), @joycethefairy (BC value = 0.671).

I also ran a 'Cluster' calculation, which calculates "the number of edges connecting a vertex's neighbors divided by the total number of possible edges between the vertex's neighbors." (Hansen, Shneiderman & Smith, p16). Basically, it tries to spot the clusters of nodes that are more interconnected amongst each other than to other people. They are represented by represented by the different colours, which can be seen easier here - four major clusters are visible.


The next time I ran it was on the 10th October, in the afternoon before the event.


The top four this time are: @benjern (BC value = 1), @julesisapen (BC value = 0.834), @kennysia (BC value = 0.685), @spinzer (BC value = 0.357).


The third time was on the 15th October, the Thursday following the event.


The top four this time are: benjern (BC value = 1), @julesisapen (BC value = 0.625), @xpaxsays (BC value = 0.432), and @joycethefairy and @MyXpaX are equal in fourth place (BC value = 0.398).

• There are clearly more people, but not many more clusters here.
• Two new tweeters are prominent, @xpaxsays and @MyXpaX - they are 'corporate tweeters'.
• One interesting point is that although @joycethefairy has 1,521 followers, and @MyXpaX has only 19 followers, they have the same degree of centrality in this particular snapshot of the twitterverse. This shows how much the sample can influence the result of the 'social network' being analysed: within this sample thirteen followers of @joycethefairy and @MyXpaX tweeted 'xpaxblackberry', meaning they have the same weight in this sample. What has happened is that @MyXpaX keeps retweeting/mentioning and following tweeters who mention 'xpaxblackberry'.
• @kennysia, who was initially the most prominent and central person, has disappeared right off the graph. This must be because the archives are only kept for so long, and he has not tweeted recently enough; or that the tweets have gone beyond the 10-page limit (discussed here, I'm not sure what the exact story is). Or nodeXL only limits itself to a certain amount of days.

Conclusions
• To do an experiment like this better one would have to analyse more carefully over time (e.g. doing a search every hour or something - for a more sophisticated example see Tim Highfield's foray).
• What's interesting is to note the shifting of the centre of this particular 'conversation'.
• To get an idea of the relative importance of the tweeters, or at least assumed importance, it would be necessary to include some computation of the number of followers each one has.
• The reciprocity of follower/following is important too. The more followers there are compared to following, the more significant that tweeter is likely to be.
• The connections between tweeters are generally quite dense - that is to say, although there is clustering of smaller groups, there are lots of ties between the groups too.
• Overall, the leading tweeters are also leading bloggers. For the moment, I would say that there's no clear differentiation between the Malaysian blogosphere and twitterverse.