Stock price change prediction using news text mining

Abstract

Along with the advent of the Internet as a new way of propagating news in a digital format, came the need to understand and transform this data into information. This work presents a computational framework that aims to predict the changes of stock prices along the day, given the occurrence of news articles related to the companies listed in the Down Jones Index. For this task, an automated process that gathers, cleans, labels, classifies, and simulates investments was developed. This process integrates the existing data mining and text algorithms, with the proposal of new techniques of alignment between news articles and stock prices, pre-processing, and classifier ensemble. The result of experiments in terms of classification measures and the Cumulative Return obtained through investment simulation outperformed the other results found after an extensive review in the related literature. This work also argues that the classification measure of Accuracy and incorrect use of cross validation technique have too few to contribute in terms of investment recommendation for financial market. Altogether, the developed methodology and results contribute with the state of art in this emerging research field, demonstrating that the correct use of text mining techniques is an applicable alternative to predict stock price movements in the financial market.