AI-driven methodology for mining scientific literature and extracting databases: A case study on a chemical process
Gal Sror, Gilad Etzyoni, Michael Fire, Ronen Berkovich, Lior Rokach, Moti Herskowitz · 2026 · 2 citationsRead the paper
• Introduced a generic AI-driven methodology for constructing scientific databases. • LLM-based pipeline converts unstructured papers into structured data. • The methodology was demonstrated and validated on a case study on a chemical process. • Extraction accuracy: text and tables >99% F1 score, 85.8% for visual figures. Scientific papers are a primary source of knowledge for research and development, yet the information they contain is highly unstructured as it being spread across text, tables, and figures in inconsistent formats. Structuring this dispersed information into a coherent database is a central informatics challenge. In this work, we present a generic artificial intelligence (AI)-driven methodology to automate the construction of domain-specific databases from scientific literature. The methodology includes three Large Language Model (LLM)-based stages: identifying domain-specific literature, classifying papers by relevance, and extracting the scientific data they contain. We demonstrate and validate this approach in a case study related to chemical reactions associated with environmental issues by constructing a catalytic performance database for the CO hydrogenation process. Data from over 1,000 disparate papers were extracted and transformed into a tabular database, containing over 9,600 entries. The extraction performance proved highly effective, achieving an F1 score of >99% for text and tables, and 85.8% for figures. This methodology can be easily adapted and implemented in other fields.
Why this research matters to Seedlabs
Seedlabs @seedlabsthe platform
We're Seedlabs — a platform that turns recent research into startup ideas — so this paper sits squarely in our problem space. We stream papers and run LLM ideation, but mostly one paper at a time. This work goes further: over 1,000 papers converted into a 9,600-entry structured database, with >99% F1 on text and tables. That is the deep multi-paper synthesis on our roadmap. Queryable, structured evidence across a field would sharpen our clustering, scoring, and validation loops. Their work, not ours — but a capability we'd want to adopt.
Claim this paper supports: Supports Seedlabs' claim that recent research is systematically underused; the multi-paper LLM extraction pipeline matches its roadmap goal of deep synthesis across many papers.