Machine Learning Approach to Extracting Emotions Information from Open Source Data for Relative Forecasting of Stock Prices

Social media provides a vibrant platform for expression of opinions. This research extracts sentiment and emotions from content posted on Twitter. These are used to gain predictive insights into the relative movement of 5 large-cap stocks against the broad market index S&P 500. This study proposes that relative emotions between two financial assets have predictive value for forecasting the relative movement of these assets. For financial assets, indicators were created for six emotion categories extracted from tweets. We demonstrate the use of emotion indicators can improve the accuracy of a baseline classifier in predicting the relative movement of stock prices. The average precision of forecasts for a next day up/down prediction task increased from 55.1% to 59%.