Abstract
Sustainability remains a central global challenge, requiring a nuanced un- derstanding of how global policy frameworks align with localised priorities. However, analysing diverse data sources for sustainability assessment re- mains a key challenge, as globally issued formally structured reports often lack localised granularity, while unstructured local data lacks structure and standardisation. Existing approaches fail to systematically integrate these heterogeneous sources, limiting their effectiveness in identifying actionable sustainability insights.
This study presents an Artificial Intelligence (AI)-driven framework that leverages Natural Language Processing (NLP) techniques to integrate struc- tured and unstructured sustainability data. We applied Latent Dirichlet Al- location (LDA), BERTopic, Generative AI (GenAI), and FinBERT-based co- sine similarity to extract macroeconomic trends from formal reports—Executive Summary of IMF’s Global Stability Reports—while identifying localised sus- tainability strategies from Greenstone’s UK-based newsletters on sustainable practice. GenAI outperformed topic models in producing more coherent, di- verse, and contextually relevant topics.
To further enhance GenAI’s performance, we applied MIPROv2—a Bayesian optimisation-based prompt tuning method—which improved topic distinc- tiveness across data sources.
Our key contribution lies in aligning global and territorial sustainabil- ity discourses through AI-enhanced topic modelling. The findings demon- strate an integrated methodology that connects global policy directives with region- and industry-specific insights. This approach uncovers underexplored opportunities in the social and governance dimensions of ESG, enabling data- driven and adaptable strategies. By synthesising insights across multiple data sources, this research enables policymakers, financial institutions, and indus- try leaders to bridge sustainability knowledge gaps, align local priorities with global objectives, and drive innovative, targeted solutions.
This study presents an Artificial Intelligence (AI)-driven framework that leverages Natural Language Processing (NLP) techniques to integrate struc- tured and unstructured sustainability data. We applied Latent Dirichlet Al- location (LDA), BERTopic, Generative AI (GenAI), and FinBERT-based co- sine similarity to extract macroeconomic trends from formal reports—Executive Summary of IMF’s Global Stability Reports—while identifying localised sus- tainability strategies from Greenstone’s UK-based newsletters on sustainable practice. GenAI outperformed topic models in producing more coherent, di- verse, and contextually relevant topics.
To further enhance GenAI’s performance, we applied MIPROv2—a Bayesian optimisation-based prompt tuning method—which improved topic distinc- tiveness across data sources.
Our key contribution lies in aligning global and territorial sustainabil- ity discourses through AI-enhanced topic modelling. The findings demon- strate an integrated methodology that connects global policy directives with region- and industry-specific insights. This approach uncovers underexplored opportunities in the social and governance dimensions of ESG, enabling data- driven and adaptable strategies. By synthesising insights across multiple data sources, this research enables policymakers, financial institutions, and indus- try leaders to bridge sustainability knowledge gaps, align local priorities with global objectives, and drive innovative, targeted solutions.
| Original language | English |
|---|---|
| Journal | Sustainable Futures |
| DOIs | |
| Publication status | Published (VoR) - 11 Sept 2025 |