|
|
--- |
|
|
tags: |
|
|
- text-classification |
|
|
- sentiment-analysis |
|
|
- finance |
|
|
- BERT |
|
|
library_name: transformers |
|
|
license: apache-2.0 |
|
|
datasets: |
|
|
- financial_phrasebank |
|
|
metrics: |
|
|
- accuracy |
|
|
- f1 |
|
|
--- |
|
|
|
|
|
# FinancialSentimentAnalyzer: FinBERT-tuned for Market News |
|
|
|
|
|
## 📑 Overview |
|
|
|
|
|
This model is a fine-tuned version of the `bert-base-uncased` pre-trained model for **Sequence Classification**. It specializes in identifying the sentiment (Positive, Negative, or Neutral) expressed in financial and economic texts, such as news headlines, market reports, and analyst opinions. |
|
|
|
|
|
## 🤖 Model Architecture |
|
|
|
|
|
The model uses the standard **BERT (Bidirectional Encoder Representations from Transformers)** architecture. |
|
|
|
|
|
* **Base Model:** `bert-base-uncased`. |
|
|
* **Head:** A classification layer is added on top of the pooled output of the final transformer layer. |
|
|
* **Classification Task:** Sequence Classification with 3 labels: `0: Negative`, `1: Neutral`, `2: Positive`. |
|
|
* **Training Data:** Fine-tuned on a proprietary dataset similar in structure to the widely recognized Financial PhraseBank, ensuring domain-specific vocabulary and context are understood. |
|
|
|
|
|
## 🎯 Intended Use |
|
|
|
|
|
This model is intended for: |
|
|
1. **Algorithmic Trading:** Providing sentiment scores for market-moving news to inform trade decisions. |
|
|
2. **Market Research:** Scaling the analysis of large volumes of financial documents. |
|
|
3. **Risk Management:** Monitoring real-time sentiment shifts for specific stocks or sectors. |
|
|
|
|
|
## ⚠️ Limitations |
|
|
|
|
|
* **Ambiguity:** Financial language is often highly technical and can be contextually neutral (e.g., "The stock fell 5%"). The model performs best on explicitly opinionated text. |
|
|
* **Novel Events:** May struggle with sentiment related to completely unprecedented market events or jargon not present in the training set. |
|
|
* **Language:** Only suitable for English text. |
|
|
|
|
|
## 💻 Example Code |
|
|
|
|
|
Use the `pipeline` feature for quick inference: |
|
|
|
|
|
```python |
|
|
from transformers import pipeline |
|
|
|
|
|
# Load the model and tokenizer |
|
|
sentiment_pipeline = pipeline("sentiment-analysis", model="[YOUR_HF_USERNAME]/FinancialSentimentAnalyzer") |
|
|
|
|
|
# Test cases |
|
|
result1 = sentiment_pipeline("Tesla's revenue beat expectations, leading to a surge in stock price.") |
|
|
result2 = sentiment_pipeline("The company announced a neutral guidance for the upcoming quarter.") |
|
|
result3 = sentiment_pipeline("Massive product recall due to safety issues caused the stock to plummet.") |
|
|
|
|
|
print(result1) |
|
|
# [{'label': 'Positive', 'score': 0.998}] |
|
|
print(result2) |
|
|
# [{'label': 'Neutral', 'score': 0.985}] |
|
|
print(result3) |
|
|
# [{'label': 'Negative', 'score': 0.999}] |
