2010 is already looking like it’ll be fairly busy, not least because nearly a quarter of it is gone already. Over the next twelve months, I should finish my thesis, while other projects are also being developed and carried out: I’m tutoring in a first-year unit this semester, and am currently writing up new work on the French political blog research, first outlined at IR10 last year, for both my thesis and a conference presentation.

That presentation will be in June, at the International Communication Association conference in Singapore, as a paper co-authored with Lars Kirchhoff and Thomas Nicolai from Sociomantic Labs in Germany. Where my IR10 presentation looked at the text content of blog posts, this paper will be covering the links being made, in their various guises.

As part of this work, and indeed in preparation for research into topical networks, the links made around particular events or themes, I’ve been busy looking into the more permanent/static networks created by blogroll links from sites in the sample population. As with the IR10 work, I’m using data collected by Thomas Nicolai and Lars Kirchhoff over the first eight months of 2009, with 217 political blogs, media resources, and other related websites represented in the final collected data. For this stage, I’ve taken these sites as a starting point, making a list of each blogroll out-link from each of the 217 sites as a two-column spreadsheet (host site, site linked to), and then importing the final list into Gephi for visualisation purposes.

[Because I was using a slightly older version of Gephi, I was also converting the spreadsheet into Pajek's .net format in order to import it into Gephi using Excel 2 Pajek. However, the latest version of Gephi imports .csv, with extra import options through the .gdf format too]

Having not used Gephi before (I couldn’t get it to work when I tested out visualisation options quite a long time ago), my success in testing it out was greatly aided by the Gephi team releasing a step-by-step tutorial for new users. Importing every individual link originating from the 217 sites and following each tutorial step led to something that looks rather spectacular, although doesn’t really say much:

Of course, the risk with visualisation is that too much attention is spent on the ‘pretty’ side of things, or on preparing diagrams that look impressive (or ‘sciencey’), but don’t aid the research’s argument (or even confuse it further). While the initial aim of creating a blogroll network is to help me see the groups of sites that associate with each other, trying to get a handle on how these sites in the sample relate to each other, the warnings and advice from people such as Bernie Hogan at last year’s OII Summer Doctoral Programme have stayed in the back of my mind. As such, I’ve spent a fair amount of time over the last few weeks trying to clean up the data and improve the visualisations, not from an aesthetic point of view, but so I get a clearer sense of what I’m trying to describe.

With the full list of links containing over 5000 nodes, receiving at least one in-link from one of the 217 initial sites, one of the main problems in the first visualisation is the sheer number of nodes, and the implied overimportance of sites with many out-links (especially when these sites are the only ones linking to many nodes – it leads to large groups of satellites around nodes). The next step then, as seen above, was to restrict the nodes to those sites receiving two or more in-links from the initial 217 sites. A number of loose groupings were immediately apparent (see, for example, the top-left of the diagram), and these were followed up after the next round of cleaning the data:

In the first of these two visualisations, some nodes are coloured by their affiliation to particular political parties (either by being official sites or by containing the party name/acronym in their URL). A loose grouping of sites from the Front National (brown) and UMP (blue) in particular is apparent. In the second visualisation, I located sites that were members of three different blog communities or networks, organised around different themes or beliefs. Again, there is some loose grouping – unsurprising, considering this is a blogroll-oriented network, and often sites will have links either to the main page of the group or the other members in their blogrolls – but what is most interesting is the general location of the anti-Sarkozy group Les vigilants (in pink) between the left-wing and centrist party groupings (in the first of the two visualisations). For more details and visualisations-in-progress, check out my Flickr (and look out for updates on the related paper over the next few months!). The next important step, particularly in terms of new information, is comparing the blogroll links to the topical networks, and seeing whether the same associations are in play regardless of time or topic – this will be investigated further over the next few weeks. At this stage, in particular because of its ease of use (and not being restricted to the latest version of operating system-specific software, I’ll most likely continue to work with Gephi while I work on my thesis. I’d still like to try out Prefuse though at some point, but that may have to wait until after all this work is out of the way…