Adapted Deep-Learning training methods developed by Uber Research to create a Twitter-based stock price direction classifier. This method helps rectify the disconnect often found between a sentiment of a tweet mentioning a stock and the actual stock direction. The reweighting training method helped improve the accuracy of a CNN from around 50% to around 86%.
For more details around data data and methods, please view the technical report found in this repository. The code for all the models can be found in NLP_L2RW_Model.ipynb. The data preparation and processing methods are found in Better_NLP_L2RW_Data_prep.ipynb