MechBERT: Language Models for Extracting Chemical and Property Relationships about Mechanical Stress and Strain.

ABSTRACT: Language models are transforming materials-aware natural-language processing by enabling the extraction of dynamic, context-rich information from unstructured text, thus, moving beyond the limitations of traditional information-extraction methods. Moreover, small language models are on the rise because some of them can perform better than large language models (LLMs) when given domain-specific question-answer tasks, especially about an application area that relies on a highly specialized vernacular, such as materials science. We therefore present a new class of MechBERT language models for understanding mechanical stress and strain in materials. These employ Bidirectional Encoder Representations for transformer (BERT) architectures. We showcase four MechBERT models, all of which were pretr

SUBMITTER: Kumar P 

PROVIDER: S-EPMC11863389 | biostudies-literature | 2025 Feb

REPOSITORIES: biostudies-literature