Alpha Vantage News Classification Script

This project retrieves financial news for a given stock ticker using the Alpha Vantage News Sentiment API and classifies each news article as either forward-looking or not forward-looking. The classification is done using a pre-trained FinBERT model for sentiment analysis. The project includes fallback functionality to read news from a local CSV file in case the API is unavailable.

Features

1. News Fetching: The script fetches up to 50 news articles for a specific stock ticker (e.g., AAPL) using Alpha Vantage's API.
2. News Classification: Each article is classified as either forward-looking or not, based on the article's summary using a FinBERT model.
3. CSV Output: The output file contains time_published, title, and classification fields.
4. API Fallback: If the API fails, the script loads news from a local CSV file and continues processing.

How FinBERT Classifies News

The FinBERT model is designed for financial sentiment analysis. For this project, it classifies whether news articles are forward-looking, meaning they speculate or predict future company or market performance.

The model tokenizes the text of each news article and processes it through a neural network. The sigmoid function is applied to the output logits to produce probabilities between 0 and 1 for each class. If the probability of the "forward-looking" label exceeds 0.5, the article is classified as forward-looking; otherwise, it is considered not forward-looking.

Requirements

• Python 3.x
• Transformers (Hugging Face)
• pandas
• torch
• requests

Setup

1. Clone the repository.
2. Install the required dependencies:

   pip install -r requirements.txt

3. Replace the placeholder API key (YOUR_ALPHA_VANTAGE_API_KEY) with your own Alpha Vantage API key.

Usage

Run the main script to fetch news for the given stock ticker and classify them:

python main.py

The script will output a CSV file containing the time of publication, article title, and classification.