Beth Webster (PVC Res Policy, Swinburne U) and colleagues have an ARC Linkage Grant to use data to estimate how inventive is a new idea and to estimate the directions of technological change.

They will do it with, “a neural-network machine-learning algorithm” that can read and classify, “the natural language of each patent description and claim and scientific article available in IP Australia, European Patent Register, United States Patent Office, and the Clarivate Web of Science journal articles databases.”

“Many existing classifications for ideas are based on tree-hierarchies or catalogues that have limited meaning for users. Natural language processing can avoid the reliance on such mechanical classification as it allows for large-scale text inferences in the way human derive meanings from the text,” her funding bid stated.

This will be a big deal indeed for policy makers helping “to identify technology trajectories and make reasoned estimates of the directions of technological change.” It will improve triaging patent applications, which can take six years in Australia, by identifying knowledge frontiers. And it will reveal the global flow of ideas which existing citation conventions and patent filings do not identify. “Ideas fly over borders and communities in ways that are not necessarily detectable by formal citation and do not fit neatly within hierarchical classifications. Ideas metamorphose from one technology or use to others in ways we only recognise anecdotally,” Professor Webster writes.