Collective Sentiment Mining of Microblogs in 24-Hour Stock Price Movement Prediction

Abstract

We propose a method for collective sentiment analysis for stock market prediction and analyse its ability to predict the change of a stock price for the next day. The proposed method is a two-stage process, based on the latest natural language processing and machine learning algorithms. Our evaluation shows best performance with the SVM approach in sentiment detection, with accuracy rates of 71.84/74.3% for positive and negative sentiment, respectively. The results of sentiment analysis are used in predicting stock price movement (up or down), and we found that users' activity on Stock Twits overnight positively correlates with stock trading on the next business day. The collective sentiments in after hours have powerful prediction on the change of stock price for the next day in 9 out of 15 stocks studied by using the Granger Causality test.