Deep learning model with sentiment score and weekend effect in stock price prediction

Stock market forecasting is a popular area for both investment and research. It is also challenging due to the strong noise generated by the news, government policies, and investor emotions. Emerging works show that the sentiment from news accumulated over weekends significantly affects stock prices. In this paper, we propose a deep learning framework to incorporate the sentiment from weekend news on social media to predict stock price, and then conduct a comprehensive set of popular benchmarks for comparison. Specifically, our model uses Valence Aware Dictionary and Sentiment Reasoner (VADER) and self-defined sentiment measure to extract lexical features and evaluate sentiment opinions. Then our model employs a recurrent neural network to capture potential dependency from sentiment and price-based features. Extensive experiments are implemented on stock indices and Reddit news in a high volatility period, which show that neural networks outperform all benchmarks significantly and validate the weekend effect of news on the stock market.