|
|
--- |
|
|
tags: |
|
|
- sentiment-analysis |
|
|
- finance |
|
|
- roberta |
|
|
- text-classification |
|
|
datasets: |
|
|
- FinancialMarketNewsHeadlines |
|
|
license: apache-2.0 |
|
|
--- |
|
|
|
|
|
# FinancialNewsSentimentAnalyzer |
|
|
|
|
|
## Overview |
|
|
|
|
|
This model is a RoBERTa-base sequence classification fine-tuned for analyzing the sentiment of financial market news headlines. It classifies headlines into one of three categories: **Positive**, **Negative**, or **Neutral**. The model is specifically optimized for short, impactful text snippets common in financial reporting. |
|
|
|
|
|
## Model Architecture |
|
|
|
|
|
The core architecture is based on the **RoBERTa-base** pre-trained language model. |
|
|
1. **Pre-trained Base:** RoBERTa (Robustly Optimized BERT Pretraining Approach) provides robust feature extraction from the input text. |
|
|
2. **Classification Head:** A standard linear classification layer is added on top of the pooled output of the final RoBERTa hidden layer. |
|
|
3. **Output:** The model outputs logits corresponding to the three sentiment classes: Negative (0), Neutral (1), and Positive (2). |
|
|
|
|
|
## Intended Use |
|
|
|
|
|
This model is intended for: |
|
|
* **Algorithmic Trading Signals:** Providing real-time sentiment input to inform trading strategies. |
|
|
* **Market Monitoring:** Automatically categorizing and filtering large streams of financial news. |
|
|
* **Research:** Analyzing market reaction to specific events or company announcements. |
|
|
|
|
|
### How to use |
|
|
|
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForSequenceClassification |
|
|
import torch |
|
|
|
|
|
model_name = "your_username/FinancialNewsSentimentAnalyzer" # Replace with actual hub path |
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name) |
|
|
model = AutoModelForSequenceClassification.from_pretrained(model_name) |
|
|
|
|
|
headline = "Tesla Stock Surges 8% Following Unexpectedly Strong Q3 Delivery Numbers" |
|
|
inputs = tokenizer(headline, return_tensors="pt") |
|
|
|
|
|
with torch.no_grad(): |
|
|
logits = model(**inputs).logits |
|
|
|
|
|
predicted_class_id = logits.argmax().item() |
|
|
predicted_label = model.config.id2label[predicted_class_id] |
|
|
|
|
|
print(f"Headline: {headline}") |
|
|
print(f"Predicted Sentiment: {predicted_label}") |
