The primary task in automated formatting of document elements is identifying their classes. An algorithm for classifying elements within a text document was based on Gradient Boosting on decision trees implemented in CatBoost. To enhance classification quality the following approaches have been considered: data preprocessing and feature engineering tailored to the dataset's specifics, training alternative models applicable to this task such as Logistic Regression, K-Nearest Neighbors classifier, Random Forest Classifier, as well as optimizing the hyperparameters of CatBoost and of the above models. A hypothesis regarding the sequential structure of the object space for determining an object's class based on its context is explored. This involves training an LSTM model on selected features specifically tailored for the purpose. The identification of paragraph features for these models involves graphemic, semantic, and syntactic computer analysis. The study analyzes the combination of these approaches and its outcomes. The findings can be applied to classify elements formatted according to various text document standards for subsequent automated verification of compliance with assumed requirements.