To continue the previous Twitter network analysis with Reaktor, I downloaded more data from Twitter. This time with a python script called twecoll, which produces a Gephi compatible graph file. Using this script requires setting up a Twitter app, since the script uses your app’s consumer key and secret. Running twecoll script is very straightforward:
- twecoll init [twitter handle] → Generates .dat-file with account details data (friends, followers, avatar URL, etc. for account friends), populates an img directory with avatar images.
- twecoll fetch [twitter handle] → populates the fdat directory, contains .f-files with list of friends of a user
- twecoll edgelist [twitter handle] → .gml-file with nodes and edges
The script took around 24 hours to scrape ReaktorNow‘s first and second degree relationships and their details. With Gephi the modularity of the nodes was calculated and modularity class was used to color nodes. Then nodes were resized by their degree and nodes with degree smaller than 15 were filtered out. Finally ForceAtlas2 algorithm was used for layout. As a result, the graph looked like this:
However, exploring such a large and complex graph is rather inconvenient with Gephi. By using SigmaExporter plugin, the graph can be exported as an interactive HTML5 webpage . The representation allows user to inspect modularity groups separately, search user by their Twitter handle and check detailed information of each node.
The explorable graph can be found here: https://g00g0l.github.io/g00g0l.github.io/#
By taking a closer look at the groups and their key influencers (by in-degree), some characteristics can be found.
Group 1 (green) – 344 members
A lot of active users of Twitter and media. Digitalists.
Group 2 (purple) – 495 Members
More techy group, with Reaktor, Futurice and Mikko Hyppönen of F-Secure.
Group 3 (blue) – 234 Members
International, techy and entrepreneurial. Bubbling under: Techcrunch, WIRED, Supercell, Nasa and SpaceX
Group 4 (orange) – 99 members
Based on the top profiles, many people with connections/interest to design. Many Reaktor employees and other developers.
Group 5 (red) – 45 members
Very Finnish group with IT people, the network is scattered.
Group 6 (green)
Maybe a bit more business / leadership oriented network – also scattered
Analysing the networks became more and more difficult when moving from group 1 to group 6. The first groups are more tight and more connected within the group. The latter groups are more loosely connected and therefore hard to classify.
Network table in CSV form can be found here: reaktor_data