Fine-tuned AI for tracking policy demands and studies

This Learning Note describes the development of an AI-based system using fine-tuned language models to support researchers in identifying and analyzing policy demands. The Alliance’s PISA team developed an annotated dataset from policy documents, labeling key elements such as drivers, outcomes, and...

Descripción completa

Detalles Bibliográficos
Autores principales: Yego, Francis, Song, Chun, Laporte, Marie-Angelique
Formato: Brief
Lenguaje:Inglés
Publicado: 2025
Materias:
Acceso en línea:https://hdl.handle.net/10568/175054
_version_ 1855528711363756032
author Yego, Francis
Song, Chun
Laporte, Marie-Angelique
author_browse Laporte, Marie-Angelique
Song, Chun
Yego, Francis
author_facet Yego, Francis
Song, Chun
Laporte, Marie-Angelique
author_sort Yego, Francis
collection Repository of Agricultural Research Outputs (CGSpace)
description This Learning Note describes the development of an AI-based system using fine-tuned language models to support researchers in identifying and analyzing policy demands. The Alliance’s PISA team developed an annotated dataset from policy documents, labeling key elements such as drivers, outcomes, and interventions, and classifying texts as either foresight or ex-post studies. The AI model, based on RoBERTa, performed Named Entity Recognition and classification tasks, achieving high precision for socioeconomic and biophysical entities. However, it faced challenges in distinguishing study types and interpreting nuanced contexts. The Note highlights technical and non-technical challenges, and emphasizes the importance of modular AI models and interdisciplinary collaboration for effective policy analysis. Future efforts aim to enhance context reasoning and deploy user-facing tools like web portals or chatbots.
format Brief
id CGSpace175054
institution CGIAR Consortium
language Inglés
publishDate 2025
publishDateRange 2025
publishDateSort 2025
record_format dspace
spelling CGSpace1750542025-11-05T11:58:28Z Fine-tuned AI for tracking policy demands and studies Yego, Francis Song, Chun Laporte, Marie-Angelique machine learning artificial intelligence policy analysis evaluation techniques This Learning Note describes the development of an AI-based system using fine-tuned language models to support researchers in identifying and analyzing policy demands. The Alliance’s PISA team developed an annotated dataset from policy documents, labeling key elements such as drivers, outcomes, and interventions, and classifying texts as either foresight or ex-post studies. The AI model, based on RoBERTa, performed Named Entity Recognition and classification tasks, achieving high precision for socioeconomic and biophysical entities. However, it faced challenges in distinguishing study types and interpreting nuanced contexts. The Note highlights technical and non-technical challenges, and emphasizes the importance of modular AI models and interdisciplinary collaboration for effective policy analysis. Future efforts aim to enhance context reasoning and deploy user-facing tools like web portals or chatbots. 2025-06-01 2025-06-11T09:24:21Z 2025-06-11T09:24:21Z Brief https://hdl.handle.net/10568/175054 en Open Access application/pdf Yego, F.; Song, C.; Laporte, M.A. (2025) Fine-tuned AI for tracking policy demands and studies. Learning Note No. 7 – Quantitative studies. 3 p.
spellingShingle machine learning
artificial intelligence
policy analysis
evaluation techniques
Yego, Francis
Song, Chun
Laporte, Marie-Angelique
Fine-tuned AI for tracking policy demands and studies
title Fine-tuned AI for tracking policy demands and studies
title_full Fine-tuned AI for tracking policy demands and studies
title_fullStr Fine-tuned AI for tracking policy demands and studies
title_full_unstemmed Fine-tuned AI for tracking policy demands and studies
title_short Fine-tuned AI for tracking policy demands and studies
title_sort fine tuned ai for tracking policy demands and studies
topic machine learning
artificial intelligence
policy analysis
evaluation techniques
url https://hdl.handle.net/10568/175054
work_keys_str_mv AT yegofrancis finetunedaifortrackingpolicydemandsandstudies
AT songchun finetunedaifortrackingpolicydemandsandstudies
AT laportemarieangelique finetunedaifortrackingpolicydemandsandstudies