High-Throughput Extraction of Phase–Property Relationships from Literature Using Natural Language Processing and Large Language Models
Name
40192_2024_Article_344.pdf
Size
1.93 MB
Format
Adobe PDF
Checksum (MD5)
47865c8cff056832f37d852a1a8252aa
Author(s)
Montanelli, Luca
Venugopal, Vineeth
Olivetti, Elsa A.
Latypov, Marat I.
Date Issued
March 19, 2024
Journal
Integrating Materials and Manufacturing Innovation
Publisher
Springer International Publishing
Citation
Montanelli, L., Venugopal, V., Olivetti, E.A. et al. High-Throughput Extraction of Phase–Property Relationships from Literature Using Natural Language Processing and Large Language Models. Integr Mater Manuf Innov (2024). https://doi.org/10.1007/s40192-024-00344-8
Version
Final published version
Abstract
Consolidating published research on aluminum alloys into insights about microstructure–property relationships can simplify and reduce the costs involved in alloy design. One critical design consideration for many heat-treatable alloys deriving superior properties from precipitation are phases as key microstructure constituents because they can have a decisive impact on the engineering properties of alloys. Here, we present a computational framework for high-throughput extraction of phases and their impact on properties from scientific papers. Our framework includes transformer-based and large language models to identify sentences with phase-property information in papers, recognize phase and property entities, and extract phase-property relationships and their “sentiment.” We demonstrate the application of our framework on aluminum alloys, for which we build a database of 7,675 phase–property relationships extracted from a corpus of almost 5000 full-text papers. We comment on the extracted relationships based on common metallurgical knowledge.
Subjects
Industrial and Manufacturing Engineering
General Materials Science
MIT Department
Massachusetts Institute of Technology. Department of Materials Science and Engineering
Terms of Use
Creative Commons Attribution
Persistent DSpace Link
DOI of Published Version
https://doi.org/10.1007/s40192-024-00344-8
