Scope3Trace: Evidence-Based Identification and Extraction of Scope 3 GHG Emissions from Sustainability Reports

Scope3Trace is an evidence‑grounded information extraction framework that identifies and extracts Scope 3 greenhouse gas emissions from corporate sustainability reports. It combines PDF collection, OCR parsing, LLM‑assisted page localization, table reconstruction, and a hybrid rule‑LLM extraction process with evidence verification to produce interpretable, traceable emissions data. The authors also release a multimodal dataset of organization‑level Scope 3 disclosures extracted from diverse reports, demonstrating high accuracy in retrieving Scope 1‑3 totals and category‑level details.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Computation and Language.

• llms
• computer-vision
• multimodal