Abstract
Portfolio analysis is a fundamental practice of organizational leadership and is a necessary precursor of strategic planning. Successful application requires a highly detailed model of research options. We have constructed a model, the first of its kind, that accurately characterizes these options for the biomedical literature. The model comprises over 18 million PubMed documents from 1996–2019. Document relatedness was measured using a hybrid citation analysis + text similarity approach. The resulting 606.6 million document-to-document links were used to create 28,743 document clusters and an associated visual map. Clusters are characterized using metadata (e.g., phrases, MeSH) and over 20 indicators (e.g., funding, patent activity). The map and cluster-level data are embedded in Tableau to provide an interactive model enabling in-depth exploration of a research portfolio. Two example usage cases are provided, one to identify specific research opportunities related to coronavirus, and the second to identify research strengths of a large cohort of African American and Native American researchers at the University of Michigan Medical School.
Highlights
Background & SummaryPortfolio analysis is a common practice in the finance world where options are well defined
The most recent versions of the Leiden Ranking6, developed by the Centre for Science and Technology Studies (CWTS) at Leiden University, are based on a model of science that consists of 4,535 document clusters7 partitioned from the citation network
The cluster-level analytics enabled by our model are an important addition to the type of data provided by PubMed Knowledge Graph (PKG), enabling both macro- and micro-level analysis of the research landscape
Summary
Portfolio analysis is a common practice in the finance world where options (e.g., stocks, bonds) are well defined. The most recent versions of the Leiden Ranking, developed by the Centre for Science and Technology Studies (CWTS) at Leiden University, are based on a model of science that consists of 4,535 document clusters (referred to as micro-level fields) partitioned from the citation network While these previous works made use of subscription-based citation databases (Scopus and the Web of Science), the goal of this work was to create a accurate model based on the (openly available) PubMed literature for strategic decision-making in biomedical research. The open database and tool contain detailed information that can be used to search and explore topics related to biomedical science, and to analyze these topics within the context of funding, industrial application, clinical application, translational potential, and other features This model is complementary to the recently published PubMed Knowledge Graph (PKG) which contains document level information from PubMed and other sources such as extracted bioentities, disambiguated authors and institutions. The cluster-level analytics enabled by our model are an important addition to the type of data provided by PKG, enabling both macro- and micro-level analysis of the research landscape
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
