Task: Sentence-level classification for Environmental (E), Social (S), and Governance (G) categories.
Base model: roberta-base, fine-tuned with class-weighted loss to address label imbalance.
Repository: GitHub → Green-Guard Project

📊 Metrics (Test Set)

Metric	Value
Accuracy	0.81
Macro F1	0.81
Weighted F1	0.81

Metrics computed on the same held-out test split (data/processed/splits/)
using reports/category_metrics_v1.json

🧩 Labels

{ "0": "E", "1": "S", "2": "G" }

🚀 Quick Inference

You can load and run the model directly:

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_id = "salitahir/roberta-esg-category-green-guard-v1"
tok = AutoTokenizer.from_pretrained(model_id)
mod = AutoModelForSequenceClassification.from_pretrained(model_id).eval()

text = "We invested in renewable energy sources for our operations."
inputs = tok(text, return_tensors="pt", truncation=True)
pred = torch.softmax(mod(**inputs).logits, dim=-1)
label_id = pred.argmax(-1).item()
label = mod.config.id2label[str(label_id)]
print(label, float(pred[0][label_id]))

✅ Expected output:

E 0.88

🧠 Intended Use

This model acts as Stage 2 in the Green-Guard pipeline — categorizing sustainability-related sentences identified by the Relevance Model.

⚖️ License

MIT License — open for research and commercial reuse with attribution.