Sentiment Analysis on Sustainability Reporting
summary
Uses BERT base for transfer learning to classify sentence-level sentiment in corporate sustainability reports, creating a labeled corpus of 8000+ sustainability sentences and showing the model detects positively framed negatives that off-the-shelf classifiers miss. Reports require OCR-filtering and topic segmentation first, labels are skewed toward neutral so sampling and human agreement checks matter, and evaluation shows BERT approaches inter-annotator consistency and produces per-report and peer-benchmarked sentiment scores that support quantitative balance assessments.
AI-written summary · about
chapters (8)
- 0:00KPMG sustainability project intro
- 2:12Project structure and data extraction
- 3:58Topic modeling for reports
- 7:14Sentiment in corporate reports
- 11:36BERT for sentiment analysis
- 14:52Training BERT on imbalanced labels
- 17:43BERT vs human annotators
- 19:30Key insights and applications
related talks
39mSentiment Analysis, its techniques and applications
32mUnpaired Sentiment-to-Sentiment Translation: Using Reinforcement Learning
01h 26mBagging to BERT- A tour of applied NLP
23mTarget-Based Sentiment Analysis with T5
41mUsing AI to Build Fair and Equitable Workplaces
43mSentiment Analysis in Action: Building Your Real-time Pipeline
01h 23mBeyond Sentiment Emotion Mining with Python and machine learning
01h 31mAn Introduction to Sentiment Analysis of...