Evolution of the degree distribution of a graph based on Twitter through time

> Maxime Vanmeerbeck

Evolution of the degree distribution

Evolution of the degree distribution

This visualisation is based on the dataset compiled by a team of researchers during the discovery of the Higgs boson on Twitter. The dataset describes the activity about the discovery on the social network of the users (reply, mention, retweet) over a period of several days.

In order to study the propagation of the information, the activity was modeled as a graph evolving with time : the users as nodes and the interactions as timestamped links between two users.

A common feature to describe the aggregated graph of interactions is to calculate the degree distribution of the nodes. Overall, and as expected, the final aggregated graph appears to be a scale-free network (with exponent γ=2.2). But is it the case at any time ? If not, is there a trigger ?

The goal of this visualisation is to highlight the hourly evolution of the degree distribution of the aggregated graph. As the degree distribution needs already two dimensions, I had trouble to visualize the evolution of the distribution. Eventually, the most efficient and natural solution was to use time as a third dimension to capture the evolution and get a feeling of what is going on.

I didn’t expect the results to go that way : the aggregated graph is always a scale-free network and the distribution exponent γ keeps the same value throughout the period. It even decreases a little bit. Around ⅔ of the video, the burst of activity corresponds to the real announcement of the discovery on Twitter.

The original paper of this study showed that 4 periods corresponding to events (rumors, etc) occurred during the experience. So this method could be a good way to sense patterns before drilling into the data.

This entry was posted in Plots