FinBERT Fine-Tuned on Financial Sentiment (Financial PhraseBank + GitHub Dataset)

📌 Model Description

This model is a fine-tuned version of FinBERT (ProsusAI/finbert) trained for financial sentiment classification.
It can classify financial text into three categories:

Negative (0)
Neutral (1)
Positive (2)

📂 Dataset Used

This model was trained on:
✅ Financial PhraseBank - A widely used financial sentiment dataset.
✅ GitHub Generated Sentiment Dataset - An additional dataset to test the model.

⚙️ Training Parameters

Parameter	Value
Model Architecture	FinBERT (based on BERT)
Batch Size	8
Learning Rate	2e-5
Epochs	3
Optimizer	AdamW
Evaluation Metric	F1-Score, Accuracy

📊 Model Performance

Dataset	Accuracy	F1 (Weighted)	Precision	Recall
Financial PhraseBank (Train)	95.21%	95.23%	95.32%	95.21%
GitHub Test Set	64.42%	64.34%	70.52%	64.42%

🚀 Intended Use

This model is designed for:
✅ Financial Analysts & Investors to assess sentiment of financial sentences in ex. reports, news, and stock discussions.
✅ Financial Institutions for NLP-based sentiment analysis in automated trading.
✅ AI Researchers exploring financial NLP models.

⚠️ Limitations

⚠️ May not generalize well to datasets with very different financial language.
⚠️ Might require fine-tuning for specific financial domains (crypto, banking, startups).

📥 Usage Example

You can use the model via Hugging Face Transformers:

from transformers import AutoModelForSequenceClassification, AutoTokenizer

model_name = "Driisa/finbert-finetuned-github"

# Load model and tokenizer
model = AutoModelForSequenceClassification.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Example input
text = "The company's stock has seen significant growth this quarter."

# Tokenize and predict
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=128)
outputs = model(**inputs)

# Get predicted class
predicted_class = outputs.logits.argmax().item()
print(f"Predicted Sentiment: {['Negative', 'Neutral', 'Positive'][predicted_class]}")

Loading…

Parameter

Value

Model Architecture

FinBERT (based on BERT)

Batch Size

Learning Rate

2e-5

Epochs

Optimizer

AdamW

Evaluation Metric

F1-Score, Accuracy

Dataset

Accuracy

F1 (Weighted)

Precision

Recall

Financial PhraseBank (Train)

95.21%

95.23%

95.32%

95.21%

GitHub Test Set

64.42%

64.34%

70.52%

64.42%

📥 Usage Example

You can use the model via Hugging Face Transformers:

from transformers import AutoModelForSequenceClassification, AutoTokenizer model_name = "Driisa/finbert-finetuned-github" # Load model and tokenizer model = AutoModelForSequenceClassification.from_pretrained(model_name) tokenizer = AutoTokenizer.from_pretrained(model_name) # Example input text = "The company's stock has seen significant growth this quarter." # Tokenize and predict inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=128) outputs = model(**inputs) # Get predicted class predicted_class = outputs.logits.argmax().item() print(f"Predicted Sentiment: {['Negative', 'Neutral', 'Positive'][predicted_class]}")