People use classification for objects organization into groups since ancient times. In one of his articles Robert Sokal notes that classification is high level of intellectual activity and it helps to understand the nature. Clustering is result of software algorithms applying to classification. This approach allows deploying data mining to classified information. The article describes an algorithm for a cluster characteristic element selection and its formal requirements definition. One of areas for the algorithm’s applying is intellectual text search systems. A main purpose of the article is description of an algorithm for characteristic element selection. The algorithm should have less asymptotic estimate operating time than enumeration of all elements. A main idea based on the classical method of branches and borders. An original part of the algorithm is errors estimates comparison for selected characteristic element. Also, the article describes two algorithms for random test data generation. Showed results of these tests illustrate and explain advantages of the main algorithm in comparison with the enumeration algorithm. An empirical assessment of the proposed algorithm convergence demonstrates its better efficiency. We plan to use the article results in intellectual text search area. Clustering and neural networks are main approaches used in this area.

В статье предложен алгоритм выбора характеристического элемента множества в задаче кластеризации, разработан эмпирический критерий сходимости предлагаемого алгоритма. Также предложены два алгоритма генерации исходных данных для проведения экспериментальных исследований по сравнению результатов работы и быстродействия предлагаемого алгоритма, и на их основе проведено сравнение с алгоритмом полного перебора.