Work

Do grant abstracts signal emerging research areas? Grounded entity extraction over ten years of NSF and UKRI computing awards

Year: 2026
Type: preprint
Abstract: Do the research areas named in grant abstracts grow before they appear more widely in publications? This study examined 23,684 NSF and UKRI computing awards from 2014 to 2023. A language model extracted 384,832 entities that could be matched verbatim to award abstracts. On 186 abstracts, exact span-and-type agreement with two independent AI annotators was F1 0.57–0.59, compared with 0.62 between the annotators. Terms such as federated learning and digital twin appeared in awards after being absent from the early observation window. At the broader level of OpenAlex Topics, growth in a topic's share of awards did not predict subsequent growth in its publication share beyond the existing publication trend (NSF and UKRI both b = +0.010, p = 0.33 and 0.54, respectively). The estimated two-year lead of NSF funding over UKRI funding was uncertain in the planned test (b = +0.15, Holm-adjusted p = 0.14). It was stronger in a post hoc analysis counting each distinct abstract once (Holm-adjusted p = 0.005). Follow-up analyses found closer agreement with agent labels for verbatim extraction than for a reimplementation of an earlier method, whose output was less constrained. For terms new to the grant record, grant growth preceded publication growth (b = +0.15, one-sided p = 0.04), with little improvement in forecasting. Pandemic-related awards also peaked before pandemic-related publications. Grant abstracts identify emerging terminology, but the topic-level analysis found no evidence that award trends anticipate publication trends.
Language: English
Open Access status: green
OpenAlex ID: W7236815343
Cites:
Cited by:
Related to: 10