sssssssssssssssss

This study tackles the challenge of extracting complex battery data from scientific papers using large language models. To improve both efficiency and accuracy, the authors propose a lightweight dual-model framework: one model identifies and classifies battery knowledge (BKRC), while the other extracts key information (BKIE). This approach enables the creation of a high-quality dataset with 3,043 battery performance records from 4,758 papers, cutting processing time and token usage by over 30%. The work supports a shift from trial-and-error to data-driven battery material design.