Multi-Modal AI for ESG Sentiment Analysis
Turn ESG text into truth: sentiment, stance, and greenwashing risk—powered by multi-modal AI
Skills you will gain:
About Program:
Build an end-to-end ESG AI pipeline in Python—from sustainability reports and climate disclosures to transformer-based sentiment/stance models, greenwashing risk detection, and ESG scoring, with an intro to multi-modal fusion using numeric ESG indicators.
Aim: Build practical AI models in Python to analyze ESG disclosures—sentiment/stance, greenwashing detection, and ESG scoring (including basic multi-modal fusion).
Program Objectives:
-
Build a Python pipeline to extract and structure ESG text from reports/disclosures.
-
Fine-tune transformer models for ESG sentiment/stance and climate disclosure signals.
-
Detect potential greenwashing using language + consistency indicators.
-
Create an interpretable ESG scoring prototype (text-only + optional KPI fusion).
-
Validate results with solid evaluation and explainability methods.
What you will learn?
📄 Day 1 — Hands-On ESG Data Pipeline & Labeling
- Focus: Building a clean, structured ESG text dataset from real disclosures
- Hands-On Activities:
- Collecting sustainability reports, climate disclosures, and CSR text sources
- PDF-to-text extraction, cleaning, and section segmentation (E/S/G)
- Designing a label schema: sentiment, stance, risk language, and materiality
- Creating a small labeled training set and annotation guidelines
- Packaging the dataset into a reusable notebook pipeline
🤖 Day 2 — Hands-On Transformer Modeling for ESG Sentiment & Stance
- Focus: Training NLP models that capture ESG sentiment, stance, and disclosure signals
- Hands-On Activities:
- Building baseline models (TF-IDF + traditional ML) for benchmarking
- Fine-tuning BERT/FinBERT-style models for ESG sentiment and stance classification
- Handling class imbalance and domain shift in ESG language
- Running error analysis and iterative model improvements
- Saving trained models and creating an inference notebook for new reports
🕵️ Day 3 — Hands-On Greenwashing Detection & Multi-Modal ESG Scoring
- Focus: Detecting misleading ESG narratives and building interpretable scoring models
- Hands-On Activities:
- Engineering greenwashing indicators (vagueness, promotional language, missing metrics)
- Running consistency checks across sections and time for disclosure reliability
- Creating a greenwashing risk score prototype (rules + ML features)
- Multi-modal fusion: combining text signals with numeric ESG KPIs (e.g., emissions, energy, safety)
- Building an interpretable ESG scoring model and packaging the capstone notebook deliverable
Mentor Profile
Fee Plan
Get an e-Certificate of Participation!
Intended For :
- Undergraduate, postgraduate, Ph.D. scholars, and early-career researchers interested in ESG, sustainability, or AI/ML
- Faculty members and academicians working in sustainable finance, climate risk, or corporate governance research
- Industry professionals from ESG teams, sustainability consulting, risk analytics, finance, or compliance
- Data scientists/analysts with basic Python knowledge and interest in NLP/transformers
Career Supporting Skills
Program Outcomes
- Build an ESG text pipeline (reports → structured dataset)
- Train transformer models for ESG sentiment/stance
- Prototype greenwashing risk detection
- Create an interpretable ESG scoring model (optional KPI fusion)
- Validate with evaluation + basic explainability
- Deliver reusable notebooks + capstone project