Multimodal Mining Extracts 13,740 XAS Spectra from Battery Literature
A new study introduces a multimodal literature mining pipeline that digitizes X-ray absorption spectroscopy (XAS) data from figures and text in scientific articles. The pipeline identifies XAS figures, digitizes spectral curves, and links them to metadata on the measured edge and material. Applied to battery literature, it produced an open dataset of 13,740 spectra spanning 66 absorbing elements. Expert validation confirmed accurate extraction. The work transforms fragmented published data into an AI-ready resource for materials science.
Key facts
- The pipeline uses multimodal (image and text) mining.
- It digitizes XAS spectra from full-text articles.
- The dataset contains 13,740 XAS spectra.
- Spectra cover 66 absorbing elements.
- The focus is on battery literature.
- Expert validation confirmed accuracy.
- The dataset is open and AI-ready.
- The work addresses inaccessibility of published spectra.
Entities
Institutions
- arXiv