NEWS BASED SENTIMENT ANALYSIS USING TANH GREY WOLF OPTIMIZER (TGWO) FOR STOCK PRICE PREDICTION

Abstract

Research on stock market forecasting has always been important to the financial sector. Shares price forecasting plays a significant role in raising investors' interest in an organization, which has a beneficial effect on the growth of shareholders in its stock. A substantial reward might have been available if the stock price had been correctly predicted. Due to advancements in science, technology, and the market economy, more elements today affect a company's price trend compared to the past. The traditional analytical technique cannot explain the fluctuations in stock price caused by important information hidden from view. Predicting these stock markets using confidential information can be achieved by several deep-learning time series models based on RNN and its derivatives, such as LSTM and GRU. However, their performance still needs to be improved. The use of optimization techniques like GWO, has improved the accuracy of these models, and to increase the reliability of the prediction, news sentiment has been used. The study comprised three phases. In the first phase, this study proposed a Tanh Grey Wolf Optimizer algorithm to increase the efficiency & accuracy of the model. Tanh is used to reduce the infinite search space to -1 and 1. Sigmoid and tanh are similar functions, but tanh has a more extensive range and is more symmetrical around the origin, resulting in results that are not biased. This property allows the potential for Tanh to have a superior gradient, which leads to more accuracy compared to the present basic GWO and binary GWO, which use sigmoid. In the second phase, News sentiment analysis used word embeddings to increase the predictions' reliability. In the last phase, these predictions are ensemble with news sentiment scores and classify the score into five classes, i.e., Strong Sell, Sell, Hold, Strong Buy, and Buy.