Using Machine Learning Techniques for Assessing Media Text Readability

The paper addresses the problem of assessing media text readability depending on the correlation of objective and subjective text complexity. Objective text complexity refers to a set of measurable characteristics such as baseline, morphological, syntactic and lexical characteristics of a media text (a piece of news published on university websites), which is calculated automatically for each text. In total, 34 parameters are considered. Subjective text complexity score is assigned to each text by human experts who assess the clarity, structure, cohesion and coherence of the text. The correlation between subjective and objective text complexity is studied using machine learning models. The readability level is measured using machine learning methods and regression-correlation analysis. Namely, an artificial neural network and regression models are used. It is demonstrated that the use of polynomial features and lasso regularization allows to get a compact regression model of a high quality. The use of the chosen machine learning techniques made it possible to estimate the impact of distinct linguistic features on the readability level.