Update README.md

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)

README.md

@@ -7,6 +7,7 @@ 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
@@ -17,7 +18,7 @@ The model was trained on a curated dataset of 2,000 manually annotated paragraph
 
 Performance Metrics
 Evaluated using 5-fold cross-validation:
-Weighted F1 0.856 Weighted Precision 0.891 Weighted Recall0.856 AUC-ROC0.922
+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: