project history
1990's
SciTech Strategies is a small research lab with the following mission: to develop an accurate quantitative technique for identifying Kuhnian ‘scientific revolutions’ from the open source literature. This iInitial work used a document clustering methodology, developed at the Institute for Scientific Information (ISI), to identify Kuhnian research communities. News of successful applications (primarily in pharmaceutical and chemical firms) traveled by word of mouth and was the basis for the early funding of this research project.
1998
A collaboration effort with Sandia National Laboratories marked the first major improvement in community detection. While the algorithm (VxOrd or DrL) was designed to visualize complex networks, it was used to reduce the ambiguity in assigning documents to research communities.
2008
A collaboration effort with Elsevier resulted in the first use of our document clustering algorithms for the explicit purpose of developing a commercial product. After multiple generations of improvements, those efforts are now reflected in SciVal© topics – a research evaluation tool that is provided by Elsevier.
2010
Significant progress in methodology was a consequence of IARPA’s FUSE (Foresight using Scientific Exposition) program. This was a highly collaborative effort- with multiple teams trying a variety of techniques to predict scientific breakthroughs. Dozens of program officers (who were responsible for making research portfolio decisions in their respective agencies) provided critical input into how forecasting efforts should be evaluated. Unfortunately, none of the teams in the FUSE program (including SciTech Strategies) were able to achieve the minimum level of predictive accuracy.
2014
A collaboration with CSET (Center for Security and Emerging Technology) at Georgetown University led to the first published articles where the FUSE criteria for predictive accuracy was met [1] and subsequently improved upon [2,3]. SciTech was able to achieve ~40% accuracy using the 20,000 largest research communities. Roughly 80,000 smaller research communities were excluded from this analysis.
2025
The first major breakthrough in increasing predictive accuracy for the entire set of 100,000 research communities occurred in October, 2025. The current level of accuracy is roughly 70% for all 100,000 research communities. Details of the new method will be disclosed in an upcoming chapter in a book published by the National Bureau of Economic Research entitled ‘Keeping Pace with the Frontier: National Research Portfolio Analysis’.