Mining Textual Data in Croatian

Business intelligence systems find textual data a very useful source of information. Text processing algorithms and systems in English and other world languages are well developed, which is not the case with Croatian language. This paper explores the applicability of existing systems and examines optimal parameters for Croatian. The quality of input data strongly influences clustering and classification results. Experiments are significantly better run after reducing noise. The impact of input learning set size and dimensionality are also considered. Special preprocessing for Croatian language consists of morphological normalisation, a useful step towards better results. Non-croatian specialised text mining tools are also applicable.