Networks, Communities, and the Science of Science

Abstract: 

Detecting network communities, i.e. subgraphs whose nodes have an appreciably larger probability to get connected to each other than to other nodes of the network, is a fundamental problem in network science. Here I will discuss three major issues. First, I will address the limits of the most popular class of clustering algorithms, those based on the optimization of a global quality function, like modularity maximization. Second, I will critically review the process of validation, probably the single most important issue of network community detection, as it implicitly involves the concept of community, which is ill-defined. I will discuss the importance of using realistic benchmark graphs with built-in community structure as well as the role of metadata. Finally, I will show that neural embeddings can be used to efficiently detect communities.

Science of science is the investigation of science as a system, via analysis and modeling of data on scientists and their interactions. I will show that the distributions of citations of papers published in the same discipline and year rescale to a universal curve, by properly normalizing the raw number of cites. Also, I will discuss the impact of the COVID pandemic on science.