Volkova V. The fuzzy clustering methods for multi-topic text documents

Object of research is the classification and clusterization processes applied for multi-topic text documents in artificial-intelligence systems of data processing. Subjects of the investigation are adaptive fuzzy clusterization methods of multi-topic text documents. Methods of the research are based on artificial-inteligence methods. In particular, author applies methods of the artifical neural networks and genetic optimization as well as the sequential complex-method for finding extrema of multivariate functions. Theoretical and practical results of the work are solution of relevant theoretical and practical scientific problems concerning efficiency improvement of fuzzy clusterization of multi-topic text documents in sequential regime of data processing. The novelty of results is consist of following items: 1) For the first time, the model of adaptive fuzzy neural network have been proposed applying special nonlinear evaluators which are main difference from other neural networks. These evaluators allow to find levels of membership for input patterns of ducuments. 2) For the first time, recurrent probabilistic and possibilistic lerning methods for the developed self-organizing neural network is proposed. The difference of previously proposed method lies in appearance of fuzzificator which allow to find additional clusters during operation. 3) For the first time, author proposed the model of the clusterization system of multi-topics text documents which contains two adaptive self-organizing neural networks operating in pralel regime. 4) For the first time, a method of automatic clusterization of milti-topics text documents based on genetic algorithm with artificial selection is proposed. 5) Gain following development learning methods for neural self-organizing networks. New methods contain estimation procedure for the membership investigation of clusters based on the membership functions. Previous methods contain rule "winner gets all". Adoption level is determined by realizations in scientific library of KhNURE, scientific and technical library of National Aerospace university of Zhukovskiy, Kharkov Aerospace Institute, and Artificial Inteligence Department, Kharkov National University of Radio Electronics. Area of application cover organizations which provide development of artificial systems for data processing using fuzzy clusterization. Also, results can be applied in education, in the branch of intelligence data processing.