2nd Place winner of Fall 2019
by Leon Smith, Sam Nadjari, and Brady Boylan
In times of public unrest, some individuals involved in protests and demonstrations turn violent. When feeling passionately about their opinions, they might turn to social media platforms to voice their opinions. Our goal was to identify individuals that use aggressive and hateful text in order to promote violence at these local protests. We developed a two part machine learning model that classifies individuals into clusters in order to identify groups of those who might provoke violence. The first part of the model uses a sentiment analyzer to detect the mood of the speech and the second part uses tweet profile information to understand the patterns that they tweet in. Using Charlottesville Twitter data from the time of the Unite the Right protest in 2017, we were able to discern a group of individuals who provoked violence.