Vansh180/FinBERT-India-v1

🧾 Model Card — FinBERT-India-v1

🧠 Model Overview

FinBERT-India-v1 is a fine-tuned financial sentiment analysis model trained specifically for Indian financial news and stock market headlines. further adapted to understand India-specific financial context, market language, and sentiment nuances.

Trained on a curated dataset of India-focused financial articles, this model effectively captures regional market language, economic tone, and sentiment patterns, classifying each headline as positive, neutral, or negative.

It is designed to assist in financial analytics, market forecasting, and investment decision-making by providing precise sentiment insights tailored to the Indian financial landscape.

---

🏗️ Training Details

• Base Model: yiyanghkust/finbert-tone

• Framework: Hugging Face Transformers

• Training Hardware: Google Colab GPU (T4)

• Epochs: 8 (early-stopped at best validation performance)

• Batch Size: 8 (train), 16 (eval)

• Learning Rate: 3e-5

• Optimizer: AdamW

• Dataset Size: 7,451 labeled financial news samples

• Label Classes:

  • 🟢 Positive
  • ⚪ Neutral
  • 🔴 Negative

---

📊 Dataset Description

The dataset was LLM-labeled using an advanced large language model-based annotation pipeline, inspired by FinBERT’s financial sentiment framework and refined through manual quality validation. It consists of Indian financial news headlines collected from various stock market sources and business outlets.

Label	Count	Percentage
Positive	~45%	Market gains, optimism, positive earnings
Neutral	~33%	Factual statements, mixed signals
Negative	~20%	Market declines, losses, or risk sentiments

---

🎯 Evaluation Metrics

Metric	Score
Eval Loss	0.54
Accuracy	76.8%
Precision	76.7%
Recall	76.8%
F1 Score	76.4%

✅ The model generalizes well with balanced precision and recall, and shows strong performance despite diverse phrasing and tone in Indian market headlines.

---

💬 Example Usage

from transformers import pipeline

pipe = pipeline("text-classification", model="Vansh180/FinBERT-India-v1")

texts = [
    "Sensex surges 500 points as IT and banking stocks rally.",
    "Rupee falls sharply against the dollar amid global uncertainty.",
    "TCS announces leadership reshuffle; markets await further clarity.",
]

for t in texts:
    print(pipe(t))

Output:

[{'label': 'positive', 'score': 0.92},
 {'label': 'negative', 'score': 0.88},
 {'label': 'neutral', 'score': 0.73}]

---

🧩 Intended Use

• Sentiment analysis for Indian stock market news
• Financial report tone classification
• Feature extraction for stock price forecasting models
• Trend analysis in algorithmic trading pipelines

---

⚠️ Limitations

• The model is optimized for Indian market news; performance may vary on global news.
• LLM-based labeling introduces minor noise.
• Headlines containing sarcasm or ambiguous sentiment may be misclassified.

---

🧑‍💻 Developer

• Author: Vansh Momaya
• Institution: D. J. Sanghvi College of Engineering
• Focus Area: Financial AI, NLP, Data Science and Machine Learning for Indian Markets
• Email: vanshmomaya9@gmail.com

---

🌍 Citation

If you use FinBERT-India-v1 in your research or project:

@online{momaya2025finbertindia,
  author       = {Vansh Momaya},
  title        = {FinBERT-India-v1: A Domain-Specific Sentiment Analysis Model for Indian Financial Markets},
  year         = {2025},
  version      = {v1},
  url          = {https://huggingface.co/Vansh180/FinBERT-India-v1},
  institution  = {D. J. Sanghvi College of Engineering},
  note         = {Fine-tuned model for analyzing Indian financial news and stock market sentiment},
  license      = {MIT}
}

---

🚀 Acknowledgements

• FinBERT-tone — Base model
• Hugging Face Transformers — Training framework

---