Financial Sentiment DistilBERT

Fine-tuned DistilBERT model for financial news sentiment classification.

Model Description

This model classifies financial news headlines into three sentiment categories:

• Bearish (label 0) — Negative market sentiment
• Bullish (label 1) — Positive market sentiment
• Neutral (label 2) — No clear directional sentiment

Training Data

• Dataset: zeroshot/twitter-financial-news-sentiment
• Training samples: 7,504 (after stratified split)
• Validation samples: 938
• Test samples: 938

Performance

Usage

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="Riddhi2952/financial-sentiment-distilbert"
)

result = classifier("$AAPL beats earnings estimates revenue up 12%")
print(result)
# [{'label': 'Bullish', 'score': 0.99}]

Training Details

• Base model: distilbert-base-uncased
• Max sequence length: 23 tokens
• Batch size: 16
• Learning rate: 2e-5
• Epochs: 3
• Class imbalance handling: Class weights (Bearish 2.19x, Bullish 1.64x)

Limitations

• Negation handling: Complex negations like "not expected to miss" may be misclassified
• Domain specific: Trained on Twitter financial news, may not generalize to formal reports
• Short text: Optimized for headlines (avg 11.7 tokens), not long documents