FinEdgeX-Sentiment

A specialized financial sentiment analysis model fine-tuned for the Indian stock market (NSE/BSE).

Model Description

FinEdgeX-Sentiment is built on top of FinBERT and further fine-tuned with:

• Financial PhraseBank (4,840 sentences with 66%+ annotator agreement)
• Custom Indian Market Dataset (55 curated examples covering NSE, BSE, SEBI, RBI contexts)

Labels

Label	Description
BULLISH	Positive financial sentiment (growth, profit, upgrades)
BEARISH	Negative financial sentiment (losses, downgrades, risks)
NEUTRAL	No strong directional sentiment

Usage

from transformers import pipeline

classifier = pipeline("text-classification", model="Somadhi-Chaitanya-09/finedgex-sentiment")
result = classifier("Reliance Industries reports record quarterly profit of ₹19,878 crores")
print(result)
# [{'label': 'BULLISH', 'score': 0.97}]

Training

• Base Model: ProsusAI/finbert
• Framework: PyTorch + HuggingFace Transformers
• Training Data: Financial PhraseBank + Custom Indian Market Data
• Epochs: 5 (with early stopping)
• Learning Rate: 2e-5
• Batch Size: 16

Built By

FinEdgeX — AI-Powered Indian Stock Market Intelligence Platform

License

Apache 2.0