Read on Twitter
in this paper we analyzed the conspiracy across 3 main research goals. 1/n https://twitter.com/iDRAMALab/status/1353713580038889474
RQ1 was to find what content related to the conspiracy could be considered canonical. to investigate, we crawled 6 different aggregation sites. we saw (perhaps unsurprisingly) that they had high disagreement in terms of canonical content 2/n
related to canonical content, we investigated whether the main conspirator is one individual by using simple stylometric indicators... 3/n
... we found that when looking at these measures (like vocabulary richness) across tripcodes, one in particular showed a statistically significant difference - potentially indicating a different writing style 4/n
RQ2 was to investigate the actual content. for this we used a full connected graph made of word embeddings and focused on the terms closest to the graph's centroid... 5/n
we visualized the term "throw" to show its relationship to other words in the corpus and saw clear communities (themes) produced in the content. for example insurgence (red) and religion (magenta): 6/n
lastly, in RQ3 we looked at the way the conspiracy spread on mainstream online social networks through links to aggregation sites. we saw that banning large conspiracy related communities does a lot, but not everything, to stop the information flow: n/n
