ESG_BiodiversityBERT_Commitment_Specificity_Classification

Biodiversity Commitment Specificity Classifier

Model Overview: This binary text classification model evaluates the specificity of biodiversity commitments in corporate sustainability reports. Designed as a second-stage classifier for paragraphs already identified as commitments, it distinguishes between:

Specific: Commitments detailing concrete actions, measurable targets, clear strategies, or verifiable implementation plans

Non-specific: Vague, ambiguous, or unverifiable commitment statements

Model Architecture: Built on ClimateBERT, a DistilRoBERTa-based model pre-trained on climate-related text, this classifier was fine-tuned to assess commitment specificity in corporate biodiversity 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

Performance Metrics: Evaluated using 5-fold cross-validation: Weighted F1: 0.856 Weighted Precision: 0.891 Weighted Recall: 0.856 AUC-ROC: 0.922

Pipeline Recommendation: For optimal results, use this model in combination with our commitment detection model:

Stage 1: Identify biodiversity-related commitments Stage 2: Classify commitment specificity (this model)