Mechanisms of Evolution in a Co-authorship Network: The Case of Phytomedicine Publications
Rodrigo Mazorra, Than Ha Trinh, Srinidhi Vasudevan, Michael Basios, Philip Schiffer, Soong Moon KangMost studies in co-authorship and collaboration networks rely on theories of embeddedness or clustering for identification of small world structures and communities in collaboration networks. However, there is a gap in the research on the vastly ignored mechanisms of action that lead the actual structure of these networks. Here we develop and apply a new scalable methodology for the identification of the co-evolution of a co-authorship network. As a case study, we selected the network of researchers publishing on Phytomedicine on Pubmed, over a period of 27 years. Phytomedicine studies target multiple natural drugs, many of them with ethnopharmacological evidence that can be regarded as a solution to improve therapeutic efficacy and safety. This is a historically small and potentially disconnected field but predicted to become important in the near future. Thus, analyzing this network provides the opportunity to test a novel methodology to explain co-evolution of how keywords and collaboration among authors co-evolve.
Quantitative methods were used across the analysis aided by a heuristic approach of natural language processing and the use of keywords for cluster characterization. The challenge to solving the dynamics of convergence in collaboration, over the studied period of time encompassed: a) the network analysis of authors and publications, b) natural language processing, c) unsupervised machine learning to cluster keywords and d) input of an expert in biological sciences.
Our approach provides an alternative view of the underlying network mechanisms. In contrast to the mainstream view that argues that structural embeddedness lead to the small-world networks being created. We find the following three benefits of this approach. First, inclusion of the underlying mechanisms of action of these structures can explain co-evolution. Second, it is possible to identify the convergence of clusters over time. Third, models are proposed to explain the formation of a giant component and smaller clusters. Our new method has the capacity to scale beyond the subject of analysis of this study. This will allow for future research in the field incorporating the underlying mechanisms of evolution of collaboration networks structures.