A new, Battelle-developed system known as TechAware Search Continuous Automated Scanning for Technology Transformation (CASTT) proposes to help identify epidemic threats through a mix of text data analysis, behavioral change prediction, and creation of a graph neural network.
The project is being developed for a contract with the Intelligence Advanced Research Projects Activity (IARPA). As part of that effort, Battelle has focused its technology on monitoring changes in domains related to epidemics. Scientific communication, such as the peer-reviewed publication PubMed, allows for massive troves of complex information on emerging threats and opportunities. However, the pace of creation and wealth of data can make it tricky for national security officials to identify what threats demand swift response.
“This project shows significant promise for the use of graph neural networks to predict real-world, rare events using high-volume text data,” Allen Chen, Battelle lead data scientist, said. “We’re excited to enter the next phase of this work to further refine this important tool and get it into the hands of those who need it.”
CASTT scrolls the aforementioned publications while allowing the separation of new ideas from the development of scientific consensus to allow for more prompt decision-making. It assesses text through natural language processing (NLP), compiles such data into a regularly updated graph neural network, and then predicts key pandemic-related behavioral changes. Battelle’s new system already passed its first phase as well, following construction and validation by providing early and efficient warning of technological surprises to analysts.
The CASTT system will be integrated into an application. Battelle developers will work with analysts who will test the system’s effectiveness, including on Chinese sources of information.