Using financial news articles with minimal linguistic resources to forecast stock behaviour

Abstract

Stock prediction has always constituted a great challenge due to its complex and volatile nature. Most existing methods neglect the significant impact that mass media broadcasts have on the behaviour of investors. In this paper an innovative system is presented, combining information from news releases and technical indicators, in order to enhance the predictability of the daily stock price trends, and experimental results confirm the aforementioned impact. The news articles are in Modern Greek, a resource-poor language, presenting the challenge to utilise minimal linguistic resources. The impact of the number of related broadcast articles on stock prediction is estimated, and experimentation shows that too few articles may be harmful instead of helpful for capturing the investors' behaviour. A comparative evaluation against a similar prediction system, which makes on English newswire articles related to US stocks and utilises roughly equivalent text processing techniques, leads to interesting findings between the two languages.