Authorship Verification of Literary Texts with Rhythm Features

The paper is devoted to the authorship verification of literary texts of 19th-21st centuries using rhythm features and statistical analysis of these features. The authors presented algorithms that fully automatically find lexico-grammatical figures in raw texts. Then the authors calculated rhythm features basing on the frequency of the appearance and the structure of these figures. The experiments showed that many English, Russian, French, and Spanish authors were successfully verified with their rhythm features, the best F-measure for the AdaBoost classifier achieved 88-96 %. Besides, rhythm features were clearly visualized in boxplots and heatmaps that allowed to compare the text rhythm in a whole for several authors and languages simultaneously.