The article covers short text classification method based on different measures of similarity word distribution. The work presents the text classification method based on a new lexical descriptor information gain concept and a topic importance characteristic. The results of experiments are presented.