• Join your peers at SSTI's 2024 Annual Conference!

    Join us December 10-12 in Arizona to connect with and learn from your peers working around the country to strengthen their regional innovation economies. Visit ssticonference.org for more information and to register today.

  • Become an SSTI Member

    As the most comprehensive resource available for those involved in technology-based economic development, SSTI offers the services that are needed to help build tech-based economies.  Learn more about membership...

  • Subscribe to the SSTI Weekly Digest

    Each week, the SSTI Weekly Digest delivers the latest breaking news and expert analysis of critical issues affecting the tech-based economic development community. Subscribe today!

Recent Research: Predicting the commercial potential of science

May 02, 2024
By: Jerry Coughter

Traditionally, a scientific discovery's commercial potential is gauged after significant R&D. However, a recent paper by Duke University researchers Roger Masclans-Armengol, Sharique Hasan, and Wesley M. Cohen (2024) proposes a new method for assessing the commercial potential of scientific research before it's fully developed. Using a large language model to analyze scientific findings, the researchers predicted the likelihood that a discovery will lead to marketable products or processes.

A large language model is a type of artificial intelligence that analyzes large data sets to make predictions. The researchers validated their method through several means. They tested the model by comparing its predictions with a holdout sample, a common technique to validate machine learning models by splitting data into training and testing sets where some data is withheld during the model fitting process. Next, the authors analyzed how well their new method aligns with the technology transfer process at a major university, tracking how research progresses toward commercialization. Additionally, they examined how the method compares to companies' use of academic research from various universities.

The authors report three key findings:

  • the new method seems effective in predicting commercial potential—research articles in the top quartile of the dataset predicted by the model to have commercial value were 21 times more likely to be cited on patent renewals;
  • patenting university research doesn't necessarily restrict its adoption by companies; and
  • universities' and individual researcher reputations can influence (by as much as 55% of the sample) how much attention their research gets from businesses, potentially overlooking valuable discoveries from lesser-known institutions.

While the authors acknowledge refinement of the model is required and recognize the limitation of patents as an overall measure of commercial potential, this research offers insights to a promising approach to evaluating the commercial side of science. By implementing the method, researchers, corporations, and universities may be able to inform decisions about the direction of research and technology transfer potential, possibly accelerating scientific breakthroughs with real-world applications. Assuming the model is demonstrated effective across different scientific fields and contexts, it might allow decisionmakers to allocate resources more strategically, potentially leading to faster innovation and economic growth. Universities may be able to leverage this method to better understand which research has the most promise for commercialization, allowing them to focus technology transfer efforts more effectively. Additionally, while this study suggests that universities with strong reputations might get more attention from businesses regardless of the actual commercial potential of their research, a tool like this could help level the playing field and ensure valuable discoveries from lesser-known institutions aren't overlooked.

Measuring the Commercial Potential of Science: http://www.nber.org/papers/w32262.pdf

recent research