Developing forecasting models for estimating the behavior of capital markets is one of the most challenging tasks in financial decision support system research. Besides time series models, artificial neural network approaches and genetic algorithms, text mining technologies represent a promising approach to support financial decision-making. In this paper, the authors address the problem field of predicting stock price movements shortly after the publication of company-specific news. Incorporating the findings of current financial research, a two-stage text mining classification approach to forecast short term intraday stock price movements is presented. Intraday event study analyses have detected significantly different intraday stock price reactions within news sub-classes. Therefore, a forecasting approach is presented that aims at identifying those news sub-classes for which most significant price reactions can be expected. Moreover, the advantage of local classifiers over global classifiers for the most relevant subclass identified is highlighted.