Stock News Sentiment Analysis using Transfer Learning from BERT Model

This repository consists in a sentiment analysis of financial market using pre-trained BERT models

Setup

pip install -r requirements.txt

If you want to get the prediction, generate data, process it and plot, run:

python main.py True

If you just want to process and plot data already obtained, run:

python main.py False

The images generated are 24 for each company. We analyze volume transactioned and stock price (2 factors), analyzed our model vs reality, stocknews model vs reality and the two models vs reality (3 factors), smoothed and not smoothed curves (2 factors), and predicting only on the title and only on the body/corpus of the news (2 factors), totalizing 2.3.2.2 = 24

To make predictions in real time:

• Run the following file

  python get_real_time_predictions.py
  

• Access the link that appears. Sentiment with biggest value is the one predicted by the AI

Collaborators

• Mateus Nobre - https://github.com/mateusnobre
• Kenji Yamane - https://github.com/kenji-yamane
• João Sarmento - https://github.com/joaolrsarmento/