Stock Market Price movement prediction using RNN and Point-weight Sentiment

Abstract

Predicting the price of stocks has always been an exciting and challenging field for academics and investors for a long time as it helps to gain high-profit margins for investment companies, investors, and emerging advanced automated trading bots. Existing forecasting algorithms and studies on statistical models using sentiment analysis have shown promising results. However, due to the highly volatile nature of the stock market and many private and public variables that directly affect the market, it is very challenging to predict prices for extreme situations with reasonable accuracy. This study introduces a point-weight algorithm for tweets and news to gain a similar pattern as stock prices, combined with stock data and feed into the RNN network for time-series prediction. We will experiment with different mechanisms for point-weight algorithms to compare results, correlate with stock price patterns and changes while focusing on accuracy. Furthermore, we will experiment with other multivariate stocks and different architecture of RNN to find how it affects the accuracy of model training.