FinBERT-Long-Text: Smart Financial Sentiment Analysis

A Python wrapper for FinBERT designed to handle long financial documents (news articles, earnings calls, reports) that exceed the standard BERT 512-token limit.

Unlike standard chunking methods that average sentiment (diluting the signal), this tool uses a "Strongest Signal" (Max-Pooling) strategy to detect the most significant sentiment in a document, ensuring critical news isn't washed out by neutral filler text.

🚀 Features

• Automatic Chunking: Splits long text into 510-token overlapping chunks automatically.
• Strongest Signal Strategy: Instead of averaging scores (which returns "Neutral" for mixed documents), it identifies the chunk with the highest Positive or Negative confidence.
• GPU Acceleration: Automatically uses NVIDIA CUDA if available.
• Production Ready: Handles special tokens ([CLS], [SEP]) correctly for every chunk.

Required Libraries

• torch
• transformers
• numpy

📦 Installation

1. Clone the repository:

   git clone https://github.com/NadeeshaNJ/FinBERT-Long-Text.git
   cd FinBERT-Long-Text

2. Install dependencies:

   pip install torch transformers numpy

⚡ Usage

Ensure sentiment_analyzer.py is in your project directory.

from sentiment_analyzer import SentimentAnalyzer

# Initialize (Downloads model on first run)
analyzer = SentimentAnalyzer()

# Example: Long text with mixed history but positive recent news
text = """
    (Paragraph 1) Company X was founded in 1905... [Neutral History]
    (Paragraph 2) In 2010, they faced a minor lawsuit... [Negative Noise]
    (Paragraph 3) TODAY, they reported a record 50% profit increase! [Positive Signal]
"""

# The analyzer will detect the Positive signal despite the neutral/negative filler
confidence, sentiment = analyzer.analyze(text)

print(f"Sentiment: {sentiment} (Confidence: {confidence:.2f})")
# Output: Sentiment: Positive (Confidence: 0.98)