Data-Driven Technology Discovery Methodologies

We propose to develop Prediction of Emergent SCIENce & Technology (PRESCIENT), a system that mines a text corpus of scientific patents and publications to discovers emerging technologies that may impact WMD or CWMD. PRESCIENT will take as input a diverse corpus of patents and publications, including metadata about contributing authors and organizations, and will produce as output alerts with timelines of relevant publications, authors, and organizations. The first stage in PRESCIENTs processing pipeline extracts scientific terminology and related contextual features. The second stage uses efficient, scalable implementation of unsupervised machine learning algorithms to identify emerging clusters of patents and publications with related extracted terms that may signal an emerging technology. The third stage applies supervised machine learning to quantitatively estimate momentum, a key indicator of a technologys long-term (two to 10-year) impact potential. A final explanation stage uses summarization techniques to assemble an alert (summary) and timeline (drilldown) of salient documents and contributors for each emerging technology.