HERMES: a multi-agent framework for structured knowledge extraction from ultra-long documents in geoscience

2026-08-14

AI Summary

This study addresses the challenges posed by unstructured content and complex layouts in lengthy geoscience literature that impede knowledge computing. We propose a scalable multi-agent framework integrating text and chart parsing to establish a unified document-level extraction pipeline incorporating domain constraints, validation rules, and evidence tracing for cross-domain transferability. Experiments demonstrate that the system extracted over 30,000 entities and 450,000 attributes with an F1 score of 0.90, achieving a sixfold efficiency improvement. Furthermore, we released a cross-domain validated database, effectively resolving the difficulties associated with structured knowledge extraction from long-form scientific documents.

Abstract

Authoritative scientific knowledge in geoscience remains largely trapped in legacy monographs and historical literature, where unstructured text and complex layouts hinder computational access. We introduce HERMES, a scalable multi-agent framework that extracts structured data from ultra-long scientific documents. Using a coordinating large language model, HERMES integrates domain constraints, validation rules and evidence tracing within a unified document-level extraction process that incorporates parsed text, tables, figures and captions. Applied to the 55-volume Treatise on Invertebrate Paleontology, the system produced a structured database of 32,277 fossil taxonomic entities and 451,878 attributes, released online at https://treatise.geolex.org. Extraction performance remained stable across fossil groups (average F1 scores of approximately 0.90 for entities and 0.91 for attributes), improving per-volume efficiency approximately sixfold relative to the tested fully manual baseline. Evaluation in palaeomagnetism and geochemistry, conducted without additional model training, demonstrated transfer across distinct geoscience domains. This work provides a practical pathway to transform historical scientific literature into FAIR-oriented structured data, offering a sustainable infrastructure for data-intensive disciplines and large-scale knowledge integration.

Problem

• structured knowledge extraction
• ultra-long documents
• geoscience
• unstructured text
• FAIR data

Innovation

• Multi-agent framework
• Structured knowledge extraction
• Ultra-long documents
• Evidence tracing
• Cross-domain transfer