Click here if you prefer to read this post on Medium.com (5 minute read). Click here to look at the data visualization only on visualizing.org.
In this post I have rekindled one of my earlier data analysis and visualization projects from last year, about my explorations of conflict and insurgence dynamics using data from the GDELT event dataset and a simple epidemiological SIR model. The data visualization was done in Matlab, so it is a bit chunky, but please go ahead and check it out here.
Last year, I have started writing a paper about my the results of my exploration, but it is not ready yet. Meanwhile, here are the brief findings.
While the social dynamics that may drive social unrest events have been extensively studied recently and the general patterns regarding the distribution of event-sizes and timings are well-known, I tried to delve deeper into the problem and attempt to gain an insight into individual event dynamics. Using an event classification based on news reports from the Global Database of Events, Language and Tone (GDELT), I looked at social unrest events of different types across different scales and timelines and find that there is an underlying repetitive pattern in their dynamic. Using this information, I postulated a simple SIR system dynamics model and simulated it for various types of social unrest for the period covered by GDELT, including all armed conflicts and major protests between 1979 and 2014. I found that the great majority of unrests are characterized by very similar diffusion and decay rates, independent of their place, time or duration, thus implying a scale-free structure. What is even more interesting is that the variation of these parameters is also small when comparing across different unrest types, such as conflicts and nonviolent protests. The Achilles-heel of the analysis is the establishment of the correlation between actual events and the news reports covering them, for which there is limited literature. I tried to demonstrate the validity of this conjecture through semantic mining in Wikipedia and the BBC Country Profiles databases. So far, I found that there might be a possible universality in the dynamics and this could offer extensibility to dynamics disaster relief programs or social gatherings.