FinBERT for Financial Sentiment Analysis
This is a bert-base-uncased model fine-tuned on the Financial PhraseBank dataset for sentiment analysis of financial news.
This model was trained as part of the project available at https://github.com/jp-marques/financial-sentiment-analysis.
Model Description
This model is a fine-tuned version of ProsusAI/finbert for financial sentiment analysis. It classifies financial headlines into three categories: positive, negative, or neutral. The model achieves an accuracy of 80.8% on the test set, with a particularly strong F1-score of 50% for the negative class, which is crucial for risk detection.
Intended Uses & Limitations
This model is intended for analyzing the sentiment of financial news headlines, articles, and social media posts. It can be used to generate trading signals, assess market sentiment, or perform risk analysis.
Limitations:
- The model is trained on English financial news and may not generalize well to other languages or domains.
- The training data is from a specific time period and may not capture the nuances of new financial jargon or market conditions.
How to Use
You can use this model directly with a pipeline from the transformers library.
from transformers import pipeline
classifier = pipeline('sentiment-analysis', model="peejm/finbert-financial-sentiment")
text = "Stocks are soaring after the positive earnings report."
result = classifier(text)
print(result)
# [{'label': 'positive', 'score': 0.99...}]
Training Data
The model was fine-tuned on a Kaggle dataset, which contains sentences from financial news categorized by sentiment.
- Source: Financial Sentiment Analysis on Kaggle
- License: CC0 1.0 Universal Public Domain Dedication.
The data was split into training, validation, and test sets as detailed in the EDA notebook of the original project.
Training Procedure
The model was fine-tuned using the Trainer API from the Hugging Face transformers library. The hyperparameters and training script can be found in the notebook 03_finetune_finbert.ipynb in the project repository.
Evaluation Results
| Metric | Value |
|---|---|
| Accuracy | 80.8% |
| F1 Score (Negative) | 50% |
These results represent a significant improvement over baseline models, as detailed in the project's main README.md.
- Downloads last month
- 1