Jan Chołoniewski, Janusz A. Hołyst, Julian M. Sienkiewicz, Gregor Leban, Robert M. Paluch
An abundance of online news outlets enables a large scale statistical analyses of news circulation. Using the EventRegistry system (www.eventregistry.org), we made an attempt to describe and model phenomena observed in a global news sphere (such as heavy-tailed distributions of events coverage or publishers activity).
Our results indicate that a number of news items mentioning given keyword published by observed news sources follows the Temporal Fluctuation Scaling (TFS) law with two regimes. We show that analysis of scaling exponents provides information about coverage of selected concepts and entities, and a publisher’s position in an activity mean-variance log-log plot allows to extract its’ writing policy toward the concepts (stable vs. reactive).
A possible usage of epidemic models (namely, SIR) as a null model has been studied showing that certain mechanisms of news industry can be modelled this way. Moreover, results of SIR model simulations on random graphs and Barabasi-Albert networks also follow TFS, although with only one regime.
The results suggest that there is underlying network of publishers observing and copying each other. We propose a possible way to reverse engineer it, and analyse its structural characteristics.