A Textual Sentiment Analysis for Stock Price Volatility Forecasting using Bidirectional Long Short-Term Memory

A stock market is a place where business stocks and economic tools are bought and sold. It is a system of allstock transactions where stockholders and dealers purchase and retail shares. Moreover, a stock price prediction is a common problem for investors and stakeholders; so, they remain uncooperative in coming forward for market stocks. This study proposes a Bidirectional Long Short-Term Memory (Bi-LSTM) model to address the existing limitations of stock prediction. Initially, the data from Yahoo Finance is fed as input for preprocessing using a Kalman Filter (KF) that processes the data and reduces noise by dividing KF into prediction and update. Then, the preprocessed data is given as input for the transformer-XLNet, which integrates auto-encoding language into forward and backward directions. Later, the data is fed into textual sentiment analysis using Naϊve Bayes (NB) techniques. Finally, the output of the NB is fed as input to the Bi-LSTM prediction, which predicts the following-day finishing stock price based on the present-day concluding amount. Thus, BiLSTM resulted in greater efficiency when compared with the existing methods like Neutrosophic Logic Sentiment Analysis (NL-SA) in terms of accuracy (97.87%), sensitivity (89.90%), f1-score (91.12%), specificity (97.90%), andMCC (20.09%) respectively.