#MH17 #Whatever: The exceptional and the mundane entangled

Team Members

Laura Canuelas, Jorge Cid, Marloes Geboers, Emillie de Keulenaar, Mayu Lida, Wouter Meys, Mariola Pagan, Giovanna Salazar, Iris Steenhout, Alexei Tsinovoi, Lisa Krieg

Introduction

This project aims to see how controversies, narratives and images related to the MH-17 crash have unfolded on Instagram, specifically, how does tragic content get mediated on an aesthetics based platform such as Instagram.

Research Questions

Methodology Research Question 1

To answer the first research question a co-hashtag analysis was executed.

Main method:

1. Scaped 17.000 instagram images with #MH17 (08-2014 until 06-2015), using the DMI Instagram scraper and the Firefox add-on DownThemAll!, and created a gefx file with the Python library Networkx.
2. Hashtag co-occurrence network analysis with Gephi, on the basis of undirected weighted graph, where hashtags featured as nodes and co-occurrences as edges. We filtered out nodes with less than 150 edges, applied the Force Atlas 2 layout and colour clustered the nodes based on modularity, and manually removed unrelated nodes.
3. Qualitative visual exploration of the network

First finding: Two big clusters of hashtags: 1) plane crash 2) soccer player

Decision: manually deleted the soccer related hashtags, in order to see more depth in the 1st cluster

Second finding: +10000 nodes, the largest part of them have below 150 connections (see graph)

Decision: filter out all nodes with less than ca. 150 connections

Refinement:

Applying the layout force atlas

Node size on the basis of node degree (highly connected nodes are bigger)

Calculating modularity at 0.8, color partition on the basis of modularity

Findings Research Question 1

We found five clusters: 1) instagram pop culture, 2) photography (malaysian local) 3) plane crash and mourning malaysian local, 4) mixed cluster: plane crash, mourning universal/international/dutch, aviation universal, 5) political issues and international community (also Russian language tags).

Weighted graph, click to zoom in

Methodology Research Question 2

We started our research with a sample of 17238 images with their corresponding metadata (ID, hashtags, users, URL, etc), scraped to a .csv file. From this we took a random sample of 1000 posts. To find a sample of a thousand random images, we used the Microsoft Excel Random Number Generator [=RAND()]. Once random numbers were each associated to an image, we had our sample. We extracted it from the data file, all the while trying to find a way to keep it connected with its metadata.

A problem, however, arose: we needed to identify the right images from our data file, but the ID of images in there did not correspond with that in the metadata. Those in the metadata were part of the URL from which the images were originally extracted. To identify our sample in the data file, then, we extracted their ID from their URLs. This was done manually by copy-pasting the URLS into Microsoft Word and erasing the webpage base of the URL, exposing the picture ID. Once we had clean names, Emilie made a folder with the sample images.

Although we created an Image Plot using saturation scales earlier in this process, the result did not show any trends and categories, and, so, proved not to be very useful.

We analyzed the hashtag clusters that our hashtag group provided with their graphs (Gephi’s) and identified five, main trends: “political”; “Malaysia”; “professional photography”; “airplanes”; and generic mourning. We did not take into account the Instagram centric cluster, as they were not particular to the theme of the tragedy (instead, they were a group of hashtags aiming to attract more followers, specific to Instagram).

We took 50 images out of every cluster, due to time concerns and the complications of coding them manually. We coded those related to the most popular hashtags we found in each cluster.

Our coding process divided our sample into text-based images, portraits (individual and collective), memes, user generated and/or intervened content and news related (journalistic/ professional) images. We completed this coding in different spreadsheets, each dedicated to a different cluster, and finally combined them into a file that Imageplot could read.

Findings Research Question 2

(Preliminary) findings relating to user engagement and image analysis

By modifying the ImagePlot script so as to add a third dimension to the plots we were able to make some plots that made more sense, in the plot underneath the size of the images is adjusted to its likes.

You’d need to zoom in though to get a better look of this third dimension. (Saturation - brightness on the axes, which leads to clustering of memes such as the ribbon).

Discussion

Another limitation was the fact that our coding of image categories within hashtag clusters could not be taken into account when making the ImagePlots due to a lack of time, this should be done to make plots more meaningful.
Plots with categories and engagement metrics should reveal which categories of images (memes, text based, portraits, professional news images) trigger user engagement and how this is cluster/topic/issue related so as to shed light on user cultures on Instagram and their engagement with certain kinds of images.

Conclusions

Current news issues, such as the Gaza war, get entangled with the tragedy as well, which says something about the ‘messiness’ of the platform, at least when it comes to the case studied here.

Our findings resonate with Lev Manovich’s claims about the Euromaidan, where the mundane and the exceptional become intertwined. This study zoomed in on the exceptional - the MH17 crash - also in the exceptional and tragic, the mundane and vague (like photography, the search for likes and followers, and completely different political current concerns) become entangled with mourning for the MH17 victims and respective political demands.

To gain more insight in the last part of our research question: what is platform specific or platform dependent about the way the mh-17 tragedy unfolded on Instagram, a comparative analysis should be pursued so as to shed light on differences and similarities in the way narratives and images surrounding mh-17 unfold on other platforms such as Facebook, Wikipedia and Tumblr.