Effect of Rate of Change of Stock Prices with News Sentiment Analysis

Mapping the variations of stock prices in the market has proven to be a challenge to, considering impact of macroeconomic factors, such as news headlines on them. In this paper, a novel framework is implemented for predicting the effect of rate of change of stock prices through news sentiment by using a standard dataset with closing stock price rates for a chosen period. Random Forest classifier to extract the sentiments from day-today news articles to identify their polarity as positive, or negative. Bi-directional Long Short-Term models are implemented to map the sliding windows of prices along with corresponding sliding windows of sentiment scores of the articles. Thus, hybrid architecture that combines a Bi-LSTM based time series prediction model with a Random Forest based sentiment analysis model is developed and compared with some of the existing state-of-the-art methodologies. The polarity of news sentiment is determined with an accuracy of 84.92% and the price prediction model performs best with 10 headlines achieving a R-squared score of 74.84%.