Integrating sentiment analysis and topic detection in financial news for stock movement prediction

Abstract

Media-expressed information in financial news are critical for stock market prediction. Nevertheless, researchers have primarily focused on the role of sentiment analysis in predicting stock returns and volatility. Here we show that topics discussed in the financial news may carry additional important information. We use a combination of sentiment analysis (using finance-specific dictionary-based approach) and topic detection (using latent dirichlet allocation) to predict one-day-ahead stock movements of major US companies. The proposed system employs a deep neural network to model complex stock market relations. We demonstrate the effectiveness of this approach compared to baselines, such as support vector machines and sentiment- and topic-based models used separately.