202005-Event-Linkage-and-Impact-Prediction

Stock Price Prediction with BERT and XGBoost using Twitter data

Report

Demo

Predicted New Price on One Day

Speed Up Animation

Prerequisites

pip install -r requirements.txt

Pipeline

• Tweet Scraping

python ./twint/twitter_scraper.py

• Feature Extraction w./ pre-processing
  • see 1. feature_generator + sentiment.ipynb
  • download the BERT model from here and place it according to the directory shown in the following command.

bert-serving-start -model_dir ./model/tmp/english_L-12_H-768_A-12/ -num_worker=1

• Feature Aggregation

  • see 2. feature_aggregator + sentiment.ipynb

• Classification Test with price-feature

  • see 3. EDA_stock + XGBoost.ipynb

• Regression Test with all features

  • see 4. XGBoost for Regression.ipynb

Setup for web application

endpoints occupied:

API	Port
feature extraction	12347
database	12346
XGBoost	12345
CORS proxy	12340

• under ./api folder

python feature_api.py

python database_api.py

python xgboost_api.py

• under ./cors-anywhere folder

node server.js

• open ./visualize/index/html to see the web app

Contributing

You can Contribute to this project with issues or pull requests.

Release Notes

See RELEASE NOTES file.

License

See MIT LICENSE file.

Contact

If you have any ideas, feedback, requests or bug reports, you can reach me at [email protected]