The LLM-Assisted I-ADOPT Variable Annotation Service

Author

Barbara Magagna

Published

March 19, 2026

The LLM-Assisted I-ADOPT Variable Annotation Service

To automate the process of variable descriptions aligned with the RDA endorsed I-ADOPT Framework, a collaboration between NFDI4Earth and the FAIR2Adapt project worked on the development of the I-ADOPT Service. The service leverages Large Language Models (LLMs) to transform natural language descriptions of observational research into I-ADOPT-aligned, machine-interpretable representations. This service enables researchers to generate FAIR-compliant metadata without requiring deep semantic or technical expertise. The resulting RDF representation is visualized in a graphical interface where users can review and edit the decomposition before publishing it as a nanopublication. A collection of variables from different domains is used as a ground truth for reference decompositions (I-ADOPT Corpus).

I-ADOPT Corpus

  • 102 variables from more than 10 domains were modelled using the I-ADOPT Visualizer
  • Decompositions were aligned and harmonized according to recurring patterns
  • All decompositions are published as issues in the examples playground repository
  • Domain experts evaluated and refined the decomposition together with semantic experts based on an evaluation schema
  • All variables are represented in RDF turtle (ttl) in this repo
  • All variables are visualized in the I-ADOPT Corpus Collection

Key Features

  1. Natural Language Input:
  • Users provide plain-language descriptions of observed variables (e.g., “surface soil temperature at 10 cm depth”).
  • The LLM processes these descriptions and decomposes them into the I-ADOPT description components.
  1. Integration with Research Platforms:
  • The service is designed to integrate with platforms like RoHub, enabling seamless adoption in research data infrastructures.
  • Supports FAIR data stewardship by generating machine-readable metadata that aligns with community standards.
  1. Community-Driven Validation:
  • The service is linked to national and European initiatives, allowing for direct evaluation and feedback from end-users and domain experts.

Our Next Steps

  • Develop Variable Design Patterns
  • Develop Decision Trees for applying the Variable Design Patterns
  • Enrich service with Patterns and Decision Trees

Applications

  • Research Data Management: Automate the creation of FAIR-compliant metadata for datasets in earth and environmental sciences.
  • Cross-Domain Interoperability: Facilitate the integration of data across disciplines by standardizing variable descriptions.
  • Semantic Enrichment: Enhance the discoverability and reusability of research data through structured, machine-interpretable annotations.

Get Started

Our service is now available for testing and integration. Whether you are a researcher, data steward, or platform developer, you can use this tool to: * Go to our I-ADOPT Service repo, follow the instructions and install the service on your computer * Generate I-ADOPT-aligned variable decompositions from natural language. * Improve the FAIRness of your datasets. * Contribute to community-driven validation and refinement of the LLM models.