File size: 1,276 Bytes
0138b73
 
 
 
 
47da5d3
0138b73
 
 
3efddfb
0138b73
 
47da5d3
0138b73
 
47da5d3
0138b73
 
47da5d3
0138b73
3efddfb
0138b73
47da5d3
0138b73
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
---
license: apache-2.0
---
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)