Stock Price Prediction Based on CNN-BiLSTM Utilizing Sentiment Analysis and a Two-layer Attention Mechanism

Abstract

The recent robust growth of the economy has instigated a heightened interest among financial experts in the domain of stock forecasting. Stock price forecasting frequently involves a non-linear time series projection due to the volatility nature of the stock market. This research proposes and develops an effective method with sentiment analysis neural network model for forecasting the closing price of the following day based on the time-series properties of stock price data. Several factors affect stock prices at the same time. Simple models can only predict with difficulty. As a result, sentiment analysis will be included in this study to increase the model's precision. The model architecture encompasses the utilization of a Convolutional Neural Network (CNN) for extracting salient features from input data, Bidirectional Long Short-Term Memory (BiLSTM) for acquiring knowledge and forecasting the extracted features, and an Attention Mechanism (AM) for capturing alterations in feature states within the time series data during the prediction process. The NASDAQ Composite Index's closing price the next day for 1281 trading days was predicted using this method in conjunction with three other methods to show the method's efficacy. The experimental results demonstrate that among the four techniques with sentiment analysis, CNN-BiLSTM-AM with sentiment analysis achieve the highest prediction accuracy and performance, and the errors of this model are the smallest. The CNN-BiLSTM-AM approach with sentiment analysis outperforms the other methods in terms of suitability for stock price prediction and is better able to guide investors towards more profitable stock investing choices.