Biodiversity Commitment Classifier
Model Overview:
This binary text classification model identifies and distinguishes biodiversity commitments in corporate sustainability reports. It classifies paragraphs as either:
Commitment: The company pledges specific actions to improve biodiversity or reduce negative environmental impacts
Non-commitment: General statements, observations, or non-actionable content related to biodiversity
Model Architecture:
Built on ClimateBERT, a DistilRoBERTa-based model pre-trained on climate-related text, this classifier was fine-tuned specifically for biodiversity commitment detection in corporate disclosures.
Training Data:
The model was trained on a curated dataset of 2,000 manually annotated paragraphs extracted from sustainability reports of Fortune Global 500 companies, ensuring high-quality labels and real-world applicability to corporate ESG analysis.
Recommended Pipeline: First use ESGBERT/EnvironmentalBERT-biodiversity to identify biodiversity-related paragraphs, then apply this model to identify commitments.
Performance Metrics:
Average of 5-fold cross-validation:
Weighted F10.928 Weighted Precision0.930Weighted Recall0.929AUC-ROC0.976
