Measuring violence: a computational analysis of violence and propagation of image tweets from political protest

Abstract

the paper explores the relation between violent images and online propagation. By applying a deep learning approach to measure the amount of violence from images shared on twitter during the protest against the 2017 G20 summit in Hamburg, Germany, the paper investigates: a) if more violent images are retweeted more than less violent images, b) if more violent images are retweeted faster than less violent images, and c) if different online groups share images representing different levels of violence. The paper finds no correlation between the number of retweets and the violence of the images being retweeted but it observes that violent images are retweeted faster than non-violent images. Moreover, the paper observes how specific subgroups share more violent images than others.

Keywords: social media, protest movement, violence