Paradigms of sentiment analysis of regional online media through the example of news websites in Yakutia

This article is dedicated to exploring the sentiment of texts from regional online media, using the example of Sakha news portals, along with comparing the levels of detail in a comprehensive model of linguistic sentiment analysis with automated sentiment analysis tools. The aim of the study is to compare the paradigms of the categories from these approaches and establish their compatibility and incompatibility for the most optimal application in the practical tasks of media monitoring. A comparison is made between the approaches of linguistic sentiment analysis and sentiment analysis, from which conclusions are drawn about the compatibility and complementarity of these approaches. The relevance of the work is determined by the growing use of automated sentiment analysis in media monitoring, their insufficient integration in working with specific regional discourse, and the necessity of bridging the computational-linguistic and traditional philological approaches to studying this category. Methods used in the research include comparative analysis, analysis by immediate constituents, discourse analysis, content analysis, full sampling, and thematic sampling. The scientific novelty of the research is defined by the fact that it is the first systematic comparative analysis of the levels of detail in automated sentiment analysis using a unified methodological framework, which allowed for the identification of patterns in the functioning of various types of sentiment in media discourse. As a result of the study, it has been established that, although automated systems successfully handle the tasks of primary classification of tonal vocabulary and determining the overall polarity of texts, they demonstrate systemic limitations. The practical significance of the work lies in testing the applicability and necessity of a multi-level model of sentiment analysis, gaps in automated analysis, and the possibility of their joint use for labor optimization. The obtained model will enable effective monitoring of the sentiments in the regional media landscape and, if necessary, detail it based on the thematic focus of the material.