The corpus consists of scientific publications downloaded in autumn 2022 using the search terms “perovskite” and “perovskites”. Any publication containing either of these terms was included in the corpus. The articles were retrieved from five sources: the arXiv API, the Elsevier Article Retriever API, the CORE API, the Springer Nature API, and web scraping of the Royal Society of Chemistry (RSC) webpages, conducted with RSC’s permission.
The articles were converted to plain text using different conversion scripts, and duplicates were removed based on matching DOIs or vector similarity. The final corpus consists of 194,322 unique plain-text articles. The full corpus collection and processing workflow is described in the following article: Sipilä, M., et al. "Question Answering Models for Information Extraction from Perovskite Materials Science Literature." Communications Materials 6.1 (2025): 260.