Stock market trend prediction using an enhanced financial news classification strategy

Abstract

Document classification, the problem of classifying documents to one or more categories, remains an ongoing field of research. One challenging sub-problem is the classification of financial news stories and the analysis of their impact on the prices of a security. Such a classification allows investors to make informed and intelligent investment decisions.

Many different types of financial products are available in the market, and different products come with different sets of news stories. As we observe that financial news in the equity market is the area with the highest demand for information, we have made this our focus. The emphasis of our study is on the automatic classification of Chinese financial news articles published in Hong Kong. Our objective is to classify these articles into good news and bad news. Good news is defined as any news stories that reflect positive changes on an underlying stock’s price, while bad news is defined as stories that reflect negative changes.

In this project, we use a traditional supervised machine learning strategy coupled with an enhanced feature selection algorithm based on both exogenous market feedback and keyword relevance score. The resulting model is then used to automatically classify financial news articles as favorable or unfavorable towards an assigned stock. Results show that a model with our enhanced feature selection mechanism will predict market movement with a higher rate of accuracy.