Sentiment-aware stock market prediction: A deep learning method

Abstract

Stock market prediction has attracted much attention from academia as well as business. However, it is a challenging research topic, in which many advanced computational methods have been proposed, but not yet attained a desirable and reliable performance. This study proposes a new method for stock market prediction, which adopts the Long Short-Term Memory (LSTM) neural network and incorporates investor sentiment and market factors to improve forecasting performance. By extracting investor sentiment from forum posts using Naïve Bayes, this paper makes it possible to analyze the irrational component of stock price. Our empirical study on CSI300 index proves that our prediction method provides better prediction performance. It gives a prediction accuracy of 87.86%, outperforming other benchmark models by at least 6%. Furthermore, our empirical study reveals evidence that helps to better understand investor sentiment and stock behaviors. Finally, this work shows the potential of deep learning financial time series in the presence of strong noises.