Data-driven characterization and analysis of fringe social networks

In this thesis we present a data-driven characterization of fringe social networks. By the term fringe social networks we mean all those small emerging structures on the Web that are not mainstream, such as Twitter or Facebook. These networks generally promote themselves as a "free-speech" alternative to the mainstream, but often serve as an incubator of misleading information, hateful and malicious content due to their lack of moderation. In particular, we will focus on the fringe social network Parler and report some statistical analysis on a dataset of 183 million Parler posts between August 2018 and January 2021. The main goal is to perform an analysis on the cascades of hashtags of related to the first impeachment of U.S. President Donald Trump. Our analysis shows how malicious and hateful trends are pumped into the network by some bad actors or some other form of manipulation. We claim that the hashtag cascade can be modeled using the Hawkes process framework with the particular choice of exponential decay kernel. We prove the goodness of our hypothesis by performing parameter estimation and present some statistical tools to evaluate the goodness of fit. Finally, our analysis allows to unveil the correlation of level of hate and misleading information to the level of attention from these social communities.