Structure dynamics of evolving scientific networks
Co-authorship networks have been extensively studied in network science as they pose as a perfect example of how single elements of a system give rise to collective phenomena on an intricate, non-trivial structure of interactions. However, co-authorship networks are, in fact, one-mode projections of original bipartite networks, which we call here scientific networks, where authors are agents connected to artifacts - the papers they have published. Nonetheless, few studies take into account the structure of the original bipartite network to understand and explain the topological properties of the projected network. Here, we create bipartite networks using extensive datasets from the American Physical Society (APS) dating back to 1893 up to 2015 inclusive, and from arXiv (1986-2015). We look at the time evolution of publications and at the dynamic structure of scientific networks considering four major features of bipartite networks, namely degree distributions, density, redundancy and cycles. We show how such features shape the formation of co-authorship networks and their observed structural properties. While the structure of the networks of most disciplines does not show significant changes over time, the appearance of large collaborations, in physics, generate largely skewed degree distributions of top nodes. The latter, in turn, induces massive cliques in the projection, triggering considerable densification of the co-authorship network, while the density of the original bipartite network remains at the same levels.
- Pub Date:
- May 2020
- Physics - Physics and Society;
- Physics - Data Analysis;
- Statistics and Probability
- 16 pages, 14 figures, 01 table