Hawkes Processes on Socialand Mass Media: : A Causal Study of the #BlackLivesMatter Movement inthe Summer of 2020

Sammanfattning: In this work we study interactions in social media and the reports in mass media during the Black LivesMatter (BLM) protests following the death of George Floyd. We implement open-source pipelines to process the data at scale and employ the self-exciting counting process known as Hawkes process to address our main question: is there a causal relation between interactions in social media and reports of street protests in mass media? Specifically, we use distributed label propagation to identify such interactions in Twitter, that supported the BLM movement, and compared the timing of these interaction to those of news reports of street protests mentioning George Floyd, via the Global Database of Events, Language, and Tone (GDELT) Project. The comparison was made through a Bivariate Hawkes process model for a formal hypothesis test of Granger-causality. We show that interactions in social media that supported the BLM movement, at the beginning of nationwide protests, caused the global mass media reports of street protests in solidarity with the movement. This suggests that BLM activists have harnessed social media to mobilise street protests across the planet. Lastly, we use Hawkes processes to model retweet cascades to examine the diffusion process in the Twitter data set, and introduce a heuristic measure of user influence.1