Методика определения возраста автора текста на основе метрик удобочитаемости и лексического разнообразия

Abstract

The article describes the approaches to determining the age of the author of an anonymous text written in Russian. The fundamental works of the subject area are considered, both proven approaches (support vector machine, naive Bayes classifier, convolutional and recurrent neural networks) and modern methods (fastText, BERT) are implemented. The study used its own data set containing 1,5 million comments from social media users. A separate experiment is devoted to assessing the impact on the classification accuracy of various text vectorization methods. As a result of a series of experiments aimed at evaluating the efficiency of the methods used and selecting informative features, a model was obtained that can predict the age of the author of an anonymous text with an accuracy of 83.2%.