Fakes, flames and memes [part 2]
Explorations on Twitter visual imagery during the 2015 no-expo demonstration in Milan
Read the first part of the story here
Matteo Azzi. After receiving his Master degree in Communication Design, Matteo worked at DensityDesign Lab as research assistant. Now he is a researcher and designer at Calibro, where he works on data-driven design projects and carries out researches in the field of data visualisation and information design.
Gabriele Colombo is a designer and PhD candidate at DensityDesign, a research lab of the Design Department of Politecnico di Milano. His research project focuses on the definition of design strategies to deal with pictorial content in the context of issue mapping online. He holds a Master Degree in Communication Design from Politecnico di Milano.
3 — Top content and specific stories
Then, we looked at top retweeted images, in order to highlight specific stories. On the left part of the visualisation, there is most retweeted content exclusively with the official hashtag in the middle part, most retweeted content with the no-expo hashtag and on the right content retweeted with both hashtags. Looking at the visualisation, the most recurring elements in all three spaces are images of fire and flames.
Beside this first quantitative approach, we tried to look closer at each top images, by using the Google Reverse Image Search tool.
When reverse searching this image of a protester with a police van on fire in the background, Google (international version) best guess is “occupy wall street violence”. And looking at the pages in which the image is used, most of them are related to the Occupy Movement indeed, together with few english pages talking about the Milan protest. This image is actually a frame extracted from a video of a riot happened in Rome on 2011.
The same process allowed us to spot out other similar cases in which images used to tweet about the Milan event were actually referring to different events.
This image, retweeted 607 times, on of the most retweeted images in the set, it is an image of different protest, happening in Turkey during the same day.
We then focused on 3 visual elements in the most retweeted images. Looking at the numbers, these are not in absolute the most retweeted images, but observing the full dataset, these three elements are the most “lively”, quickly becoming central to a visual conversation, performed through remix and appropriation practices. We identified 3 visual tropes in the most retweeted images, and then back to the full dataset in order to observe how they have been contaminated and modified. The first element we focus on, was the viral international meme of the Baltimore Mum, linking the Milan riots to the protest happened some days before in the United States. The second element -the fool- is a guy who gives a very unfortunate interview during the demonstration after the riots, and quickly become “the face” of the protest; “photo opportunity” represents an image of a tourist posing in front of a burned car in the afternoon.
It is interesting to observe how these 3 iconic elements contaminate each other in the span of one day, contributing to create different and parallel narrations worth to be explored more. As An Xiao Mina put it, these are the tools by which users appropriate the current narrative, suggesting different paths for the public conversation.
4 — Riot porn analysis through visual clustering
From all the previous analysis, the pornography of the burning car and the city set on fire emerge as the recurring elements of the day. Both from the timeline view and the most retweeted content, the burning car seems to be the most relevant visual element in both the official and the unofficial narrative. So we set out to extract from the full dataset each image depicting a burning car. Instead of using automatic content recognition, we clustered the images based on formal similarities and then used the results to ease the process of manually selecting the images representing a burning car.
We started from our full dataset, clustering images by their formal properties, then, zooming in in the areas of the space where the flames cluster, we manually selected all images actually representing a burning car. Then, clustering again the selected images, we manually grouped together similar images, based on their content.
With a mix of automatic visual clustering and manual check, we were able to group together modifications of the same image, different cropped versions and memes. For each group we created an animated composite, merging together every instance of the same shot.
Looking at the composites, it is interesting to notice that of all those burning car images, overwhelmingly occupying the visual space of the day on Twitter, most of them are actually depicting the same 2 scenes (other 4 composites are not easy to connect to any specific scene).
With this experiment, we analysed the day with different entry points, applying several levels of zoom on the same set of images. This constant movement between the close view and the distant view helped us to understand the way in which the day was perceived on Twitter.
Coming back to our title (Fakes, flames and memes), we showed how these three elements occupied the visual space of the day on both side (the official and the official), marginalising the issues raised by the protesters.
Read the first part of the story here