Community Detection in the Multiplex Network of Scientific Journal Authors

The article presents a model of a multiplex network reflecting a real scheme of collaboration of authors of a scientific journal. The initial data are extracted from the XML archive of the journal articles. The model is presented as a two-layer graph, the vertices of which correspond to the authors of the articles, and the edges — to binary relations of co-authorship and citation. The purpose of the work is to identify non-intersecting communities of authors of the network and is achieved in two stages. At the first phase, the network is reduced to the form of an undirected graph 𝐺𝑓 using “flattening algorithm”, this allows to apply the known algorithms of community detection intended for single-layer networks. So, at the second phase, two traditional clustering algorithms, Walktrap and Infomap, based on the “random walk” method are applied to the flattened network. The result of the algorithm is a set of identifiers of nodes included in the community, by which the original multilayer network structure can be restored. The comparison of the algorithms’ results is made using the 𝑟𝑎𝑛𝑑, 𝑎𝑑𝑗𝑢𝑠𝑡𝑒𝑑_𝑟𝑎𝑛𝑑 and 𝑛𝑚𝑖 indices. For 𝐺𝑓 , all indices demonstrate a high level of similarity in the algorithms’ results. The parameter 𝜌𝑐, which characterizes the degree of overlapping of layers at the community level, serves as a characteristic of multilayering. The study of this parameter showed that most communities have a zero value, i. e. the communities consist of vertices of one of the layers. It should be noted that these are the actors of the co-authorship layer, while the citation layer contains 37 % of single nodes. The results of the analysis are presented in the form of tables.