Link streams represent traces of complex systems’ activities over time, in which links appear when two system entities interact with each other; the aggregation of entities (i.e. nodes) and links is a graph. These traces have become strategic datasets in the last few years for analyzing the activity of large-scale complex systems, involving millions of entities, e.g. mobile phone networks, social networks, or the Internet.
This thesis deals with the exploratory analysis of link streams, in particular the characterization of their dynamics and the identification of anomalies over time (called events). We propose an exploratory framework involving statistical methods and visualization, with no hypothesis about data. The detected events are statistically significant and we propose a method to validate their relevance. We finally illustrate our methodology on the evolution of Github online social network, on which hundred thousands of developers contribute to open source software projects.
Clipping is a handy way to collect important slides you want to go back to later.