AI-POWERED INSIGHTS FOR ESG
Environmental, Social and Governance Sentiment Analysis on Slovene News: A Novel Dataset and Models
Paula Dodig, Boshko Koloski, Katarina Sitar Šuštar, Senja Pollak, Matthew Purver
This research introduces the first publicly available Slovene ESG sentiment dataset and a suite of models for automatic ESG sentiment detection. Leveraging large language models (LLMs) and human annotation, the study evaluates various model performances on Environmental, Social, and Governance (ESG) aspects in news articles. LLMs like Gemma3-27B and gpt-oss 20B show strong performance on Environmental (F1-macro: 0.61) and Social (F1-macro: 0.45), while fine-tuned SloBERTa excels in Governance classification (F1-macro: 0.54). A case study demonstrates the framework's ability to track company ESG perception over time.
Keywords: Sentiment Analysis, ESG, Economics, Environment, Social, Governance, Large Language Models, Dataset, Financial NLP
Key Performance Indicators
Our cutting-edge models deliver robust and nuanced sentiment analysis across critical ESG dimensions.
0.0
Environmental F1-Macro (Gemma3-27B)
0.0
Social F1-Macro (gpt-oss 20B)
0.0
Governance F1-Macro (SloBERTa)
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
ESG considerations are vital for corporate assessment, but reliable ratings are limited for smaller companies and emerging markets like Slovenia. Traditional ratings are often static, failing to capture real-time shifts in public perception and lacking diversity across news outlets. This gap is particularly pronounced in less-resourced languages due to limited data and language barriers. Our framework addresses this by leveraging LLMs and automated sentiment analysis on news articles.
Enterprise Process Flow
Feature extraction
→
Base-level classification
→
Meta-level ensemble
→
ESG predictions
550
Unique Articles in SloESG-News 1.0 Dataset
The SloESG-News 1.0 dataset, derived from the MaCoCu Slovene News dataset, was created using LLM-assisted filtering and human annotation by economics students. This resource provides sentiment labels (positive, neutral, negative, irrelevant) for Environmental, Social, and Governance aspects, making it the first publicly available Slovene ESG sentiment dataset.
| Model Family | Key Strengths | Performance on ESG Aspects |
|---|---|---|
| LLMs (e.g., Gemma3-27B, gpt-oss 20B) |
|
Generally competitive, especially where nuanced framing is less critical. |
| Fine-tuned SloBERTa |
|
Robust scores comparable to/exceeding multilingual alternatives for a highly inflected language. |
| Multilingual Models (e.g., XLM-RoBERTa) |
|
More stable but less specialized; can miss subtle Slovene-specific sentiment signals. |
| Hierarchical Ensembles (Tower A/B) |
|
Strong results overall, particularly for Environmental, indicating effective fusion of diverse models. |
The study reveals that transformer-based architectures, especially when supported by ensemble and multi-task learning, are effective for ESG sentiment classification in Slovene news. Environmental sentiment shows the strongest results, while Governance remains the most challenging, likely due to ambiguity and abstract language. Social sentiment is moderate, with multilingual models performing competitively. The small dataset size and LLM-assisted filtering are noted as limitations.
Temporal ESG Evaluation: Cinkarna Celje & Talum
A case study examining Cinkarna Celje and Talum over 15 years demonstrates the framework's ability to track dynamic ESG perception. For Cinkarna Celje, sentiment patterns closely align with regulatory decisions and remediation efforts (e.g., environmental remediation in 2017 improved E/S sentiment, EU court ruling in 2025 boosting G sentiment despite limited actual impact). For Talum, financial results and strategic restructuring (2015-2016) initially faced skepticism, but green transformation investments (2018-2019) boosted E sentiment. Notably, S sentiment across both companies increased during COVID-19 (2020) due to employment stability. The study highlights that text-based ESG sentiment primarily reflects the social construction of corporate responsibility rather than direct economic or environmental outcomes, significantly influenced by communicative and institutional factors like board changes and employee gestures.
Automated monitoring of ESG sentiment, as demonstrated by the case study, provides dynamic, fine-grained insights into corporate reputation shifts across time and media outlets. This approach has broad implications for sustainability analytics and can support interdisciplinary research by contextualizing ESG perceptions with real-world events.
Calculate Your Potential ROI
Estimate the efficiency gains and cost savings your enterprise could achieve by automating ESG sentiment analysis with our AI solutions.
Estimated Annual Cost Savings
$0
Estimated Annual Hours Reclaimed
0
Your AI Implementation Roadmap
A clear, structured path to integrate advanced AI into your enterprise, ensuring maximum impact and seamless adoption.
Discovery & Strategy
We begin by understanding your specific enterprise needs, existing infrastructure, and business goals. This phase defines the scope, identifies critical ESG data sources, and selects the optimal AI models and sentiment analysis approaches tailored to your organization.
Pilot & Integration
A proof-of-concept is developed and integrated into a sandbox environment, using a subset of your data. We fine-tune the selected models, validate initial results against your criteria, and ensure seamless technical integration with your existing data pipelines and reporting tools.
Scaling & Optimization
Upon successful pilot, the solution is deployed across your full data ecosystem. We establish continuous monitoring for model performance and data quality, provide ongoing support, and implement iterative improvements based on feedback and evolving business requirements.
Ready to Transform Your Enterprise?
Unlock the full potential of AI for deep ESG insights. Schedule a personalized consultation to discuss how our solutions can empower your business.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat