Multi-aspect analysis of emotional context for text document categorization

Abstract

The problem of texts intellectual analysis to identify their emotional context is considered. Presents the results of a review of modern approaches to the identifying emotions in texts problem, the key limitations and issues in processing complex linguistic constructions are outlined. A hypothesis is advanced: to increase the quality of digital texts categorization by emotional tone, it is necessary to account for semantic and pragmatic characteristics of their representation (sarcastic statements, emoji, contextual dependencies and other linguistic markers). A method for multi-aspect text analysis for document categorization based on identification of their emotional context is proposed. A multi-modular architecture of the method is presented, the principal component of which is a classifier built on a transformer-based language model. A description of the developed software tools using contemporary approaches is provided. The main results of computational experiments confirming the hypothesis and demonstrating the proposed approach effectiveness are discussed. Further steps for research are outlined.