Using Topic Modeling To Describe Science: International Experiences

The measurement of scientific activity has largely been confined to counting and studying publications.  Recently, however, new computational techniques have been applied to describe the scientific content of what scientific teams are funded to do and what research is being produced. This technique, known as topic modeling, provides a powerful and flexible framework for representing, summarizing and analyzing the contents of large document collections, can be used to describe research topics - and hence describe what research is being done. The topics define a simplified representation of scientific documents; the research topics are defined by the research proposal, not manual generation of taxonomies or keywords.  This approach has been applied to both NSF and NIH awarded grants, to research institutions in Australia and France, and will be applied to corpora of documents at universities in the United States.  It has a number of advantages in that it facilitates the comparison of research across different funding agencies, across countries and over time. 

This presentation will describe some of the initial results  of the approach across the US, Australia and France.