Accelerating the Discovery of Proton Conducting Electrolytes via Machine Learning‐Enabled Literature Mining

Accelerating materials innovation hinges on transforming the vast, unstructured knowledge embedded in scientific literature into structured, machine‐actionable data. Yet, existing large‐scale text‐mining frameworks are often inadequate for specialized, data‐scarce domains where high‐precision extraction from small corpora is critical. Here, we introduce a data‐efficient information extraction framework built upon a domain‐adapted bidirectional encoder representations from transformers model trained with strategically augmented datasets to accurately extract ionic conductivity values and corresponding temperature conditions for proton‐conducting ceramic electrolysis cell electrolytes. Applied to 72 previously unseen publications, the model achieves an F1 score of 0.942 and an accuracy of 0.944, yielding a structured dataset containing 176 distinct ionic conductivity entries. Analysis of this dataset uncovers key composition–property relationships, revealing that barium‐based electrolytes exhibit relatively high ionic conductivities. Guided by these insights, we identified high‐valence dopants as promising stabilizers and experimentally validated novel co‐doped compositions with enhanced chemical stability. Beyond delivering a curated electrolyte dataset, this work establishes a generalizable framework that seamlessly links automated knowledge extraction to experimental discovery, providing a scalable pathway to accelerate research in other data‐scarce areas of materials science.