Semantic Embeddings of Chemical Elements for Enhanced Materials Inference and Discovery

ABSTRACT We present a framework for generating semantic embeddings of chemical elements to advance alloy inference and discovery. This framework leverages ElementBERT, a domain‐specific BERT‐based natural language processing model trained on 1.29 million abstracts of alloy‐related scientific publications, to capture latent knowledge specific to alloys. These semantic embeddings serve as robust elemental descriptors, consistently outperforming traditional empirical descriptors across multiple downstream tasks, including predicting mechanical and transformation properties, classifying phase structures, and optimizing materials properties via Bayesian optimization. Applications to titanium alloys, high‐entropy alloys, and shape memory alloys demonstrate up to 23% improvement in prediction accuracy. Notably, machine learning models based on semantic embeddings exhibit superior performance on newly published alloy data that are not included in the training corpus, indicating promising generalization capabilities for emerging alloy compositions. Our results show that ElementBERT surpasses general‐purpose BERT variants by encoding specialized alloy knowledge. By bridging contextual insights from scientific literature with quantitative inference, our framework accelerates the discovery and optimization of advanced alloys, with potential for extension to other material classes.