Stock Price Prediction Based on LSTM Neural Network: the Effectiveness of News Sentiment Analysis

摘要

This paper retrieves news articles from the New York Times and conducts sentiment analysis for news headline and text body, then combine quantitative sentiment score with stock historical stock basic features together, using LSTM neural network to predict both future stock close price and stock return. The main purpose is to compare the prediction result of model which includes sentiment factors with that only considers historical stock basic features. LSTM neural network shows a good ability in long-term prediction, the experiment is based on LSTM model for three representative large companies in the US. The final results confirm that the prediction accuracy is higher for model that consider sentiment influence from website news article.