The next paper in this 2019 AoIR Flashpoint Symposium session is presented by Felix Münch and Ben Thies, and Cornelius Puschmann and I have also made a small contribution to it. Our project adapted an experimental algorithm to sample a language-based Twitter follower network, and this was necessary because gathering Twitter follower networks at scale has become increasingly difficult.
Information on such follower networks would open up significant new avenues for investigation that cannot be answered by examining actual interactions (via @mentions and retweets) alone. We did some such work in the QUT Digital Media Research Centre by mapping follower connections in the Australian Twittersphere as of February 2016, but this was possible only because at that time user IDs on Twitter remained consecutive and could be captured by brute force methods; today, this is no longer possible.
The present project uses a modified version of the rank degree method, therefore. This begins with a number of random Twitter accounts and then follows especially the nodes with the strongest influence (as measured by degree); in a sense, and at risk of oversimplification, this becomes a more purposive version of a snowball crawl of the Twitter follower network. The present study also filtered only for German-language accounts, in order to develop an approximate representation of the German-language Twittersphere.
The network gathered in this way focusses strongly on the most influential accounts, and so is a representation of what we might consider the backbone of the network in particular. This still lends itself to community detection and keyword extraction, however, as these accounts are likely to be highly influential for the rest of the German-language Twittersphere. This found major clusters around sports (mainly football), mainstream politics in Germany, Austria, and Switzerland, far-right politics, YouTubers, gaming, parenting, entertainment, literature, and other topics, as well as the interrelations between these groups.
