Zolotukhin O. Classification methods of multi-topic text documents using neuro-fuzzy technologіes

The thesis is devoted to methods classification of multi-topic text documents development in a sequential (online) mode under conditions of overlapping classes. The task of the multi-topic text documents classification, basic methods of document processing and existing classification methods, the main advantages and disadvantages of these methods have been discribed. A fuzzy probabilistic neural network's architecture and an adaptive neural network of fuzzy vector quantization and its learning method which provides fuzzy classification of multi-topic text documents in a sequential mode have been developed for the first time. The proposed method of neural network's learning is characterized by high speed and low computational complexity. A counter-propagation neural network's model of controlled studies has been proposed for the first time. A learning method for counter-propagation neural networks which provides a better classification for classes overlapping and increase the information processing speed has been developed.