Text Mining Systems for Market Response to News: A Survey

Several prototypes for predicting the short-term market reaction to news based on text mining techniques have been developed. However, no detailed comparison of the systems and their performances is available thus far. This paper describes the main systems developed and presents a framework for comparing the approaches. The prototypes differ in the text mining methods applied and the data sets used for performance evaluation. Some (mostly implicit) assumptions of these evaluations are rather unrealistic with respect to properties of financial markets and the performance results cannot be achieved in reality. Furthermore, the adequacy of applying text mining techniques for predicting stock price movements in general and approaches for dealing with existing problems are discussed.