Doctor Kurohashi's laboratory at Kyoto University proposed a method for classifying Japanese documents without performing any morphological parsing of these documents but just by observing the kanjis.
The model was trained using a database of texts already classified according to their topic (philosophy, architecture etc...) to extract the kanjis characteristic for each topic using a method. The kanjis found to be characteristic can thereafter be used to classify new texts depending on the kanjis observed in it.