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

Vasilev Vasilii Vasilevich ORCID: 0009-0006-2673-504X Senior lecturer; Department of English and Translation; Northeastern Federal University named after M.K. Ammosov postgraduate student; Department of German Philology; Ural Federal University named after the first President of Russia B. N. Yeltsin 677008, Russia, Rep. Sakha (Yakutia), Yakutsk, Tchaikovsky St., 30 K. 2 vv.vasilev@s-vfu.ru

Introduction

In the era of social networks and electronic media, text tonality analysis (sentiment analysis) is expanding its sphere of influence and gaining popularity. This tool is widely used in the management of recommendation sites, social networks and businesses. One of the problematic areas of tonality analysis is also the sphere of mass media. Tonality analysis is widely used today and is an important tool for monitoring sentiment in the federal media, such as the Scan system of the Interfax IG. Such services are an important tool for reputation management and monitoring of the media space or social networks. It is of interest to use such systems for regional media to identify regional sentiments, which correlates with the task of studying the perception of the local agenda by the audience ^[15]. However, in order to identify a reliable reflection of events and reactions of society, it is necessary to define a classical theoretical framework within which the validity and applicability of these tools will be considered. To do this, it is necessary to consider approaches within the framework of linguistic analysis of tonality, identify the units and categories to be analyzed, and investigate their presence in automated systems.

Problem statement

The tonality of the text, in the light of its interdisciplinarity and validity on such linguistic paradigms as communication theory, discourse analysis, and stylistics, is not in itself a rigidly determined and structured field. Thus, approaches to its analysis vary from author to author. In various works, the term is used to denote its various facets, such as the author's attitude ^[9][10], stylistic design of the text, functional load, and communicative function ^[9]. A possible solution is to create an integrative approach to the analysis of tonality.

At the same time, against the background of such a multifaceted category of tonality, modern automated sentiment analysis tools, which by their nature require structure, finiteness and determinism, widely explore the problems of tonality, encoding it in the categories of statistics, indicators and variables. There is a discrepancy in the plans of the deep symbolic, semantic, and communicative nature of the concept of tonality and the surface indicators of statistical approaches to sentiment analysis. This article aims to connect these paradigms, compare the levels of detail and build a model of interaction.

Methods and materials

For a full-fledged consideration and substantiation of the scientific framework of tonality, various works on the theory of discourse, emotivity, axiology, and evaluativeness were considered. The works of V. I. Karasik ^[9], T. V. Matveeva ^[10], and T. O. Baghdasaryan ^[5] are the mainstays for considering the category of tonality. In these works, the phenomenon of communicative tonality is considered comprehensively, from the point of view of the theory of speech acts, the theory of discourse, genre and stylistic features. Speaking about the category of tonality, it is worth referring to fundamental works on emotive vocabulary (L. G. Babenko) ^[4], functional semantics of assessment (E. M. Wolf) ^[7]. These works lay the fundamental foundation for the theoretical substantiation of the level paradigm of tonality. In addition, empirical studies of tonality in the context of local topics, such as the work of A. S. Magranov, devoted to the analysis of the tonality of messages about migrants in regional media, are important for understanding the specifics of regional media discourse ^[14].

As for the paradigms unfolding in the sentimental analysis approach, the systems under consideration will be described in the framework of fundamental works on sentiment analysis, such as B. Liu^[1], B. Pang, L. Lee^[1], and the theoretical foundations of sentiment analysis in Russian-language texts are also considered in the works of S. Smetanin ^[3].

The research material is regional media, as they provide unique subject areas in which it is possible to express locally specific ways of expressing tonality. Such texts are an ideal base for checking and testing the limitations of AI systems. The expected result would also contribute to the development of regional media and media, an understanding of the socio-political climate, and the specifics of regional sentiments. The results will help identify semantic dominants, patterns of authors' attitudes towards semantic dominants, tonal characteristics and distribution, means, and identify errors and lacunary categories of AI systems, especially in regional (characterized by a national component, location) texts.

Methods such as comparative analysis, direct component analysis, discourse analysis, content analysis, continuous sampling, and thematic sampling were used for the study.

News articles for the period from October 9 to October 24, 2025 were selected as the material. During this time frame, the media covered the topic of introducing amendments to the Constitution of the Republic of Sakha (Yakutia), which was quite resonant for the region. The selected time frame is associated with two important dates – the public hearings on the introduction of amendments on October 10 and the actual introduction of amendments on October 22, 2025. The materials were selected from media outlets of various types (according to the typology of regional media by G. S. Melnik and A. N. Teplyashina ^[11]), such as: government media (Yakutia.daily; Yakutia 24), private media (1sn.ru , Sakhaday.ru ), the sample also includes interregional media outlets that covered this issue: an expert platform "Regional comments", private news agency "Vostok.Today». This sample is based on considerations of the representation of tonality in various types of regional media. Platforms such as EurekaEgine, SentiStrength, and YesChat Sentiment Analysis Classifier were used to study automated tonality analysis systems. Table 1 describes the functionality of each of these systems.

Table 1. Functional capabilities of automated tonality analysis systems

The results of the study

A selection of articles has been made on the resonant topic for the Republic of Sakha (Yakutia) of introducing amendments to the constitution regarding education and the powers of the Head in terms of managing funds. Since the issue is socially important, the number of articles around this high-profile event is presented in sufficient volume. 50 texts have been selected, reflecting different points of view on this issue.

The research algorithm was based on various approaches to the study of tonality, which formed the basis for each of the analysis stages. The first stage is the meta–textual description, the identification of modal frameworks that determine the focus of the article, the discourse framing the communicative situation, the frame, collectively determining the motivation for choosing the means of expressing tonality. The second stage is the identification of objects of tonality (an element of reality), the identification of semantic dominants in the text. The third stage is the identification of tonality indicators at different language levels, in which the emotional and evaluative component is manifested ^[7], as well as the author's intention is expressed ^[9][10]. The fourth stage is interpretation, the relationship of these indicators of tonality with the described phenomena is established, the types of tonality are recorded ^[13]:

• by polarity – positive/negative/neutral (T. F. Bogdanova) ^[6];
• with a moderate/pronounced level of intensification (E. M. Wolf)^[7];
• according to the emotion inherent in this text (T. O. Baghdasaryan) ^[5];
• communicative type (informative, ideological, mentoring, etc.) according to V. I. Karasik ^[9];
• in terms of cooperativeness with respect to the speech genre, it is unison/dissonance (V. V. Dementyev) ^[8]. It should be noted that this set of types is important for the field of mass media, taking into account the conditions for the formation of public opinions.

The fifth stage is a comparison of the expert analysis of tonality with the analyses performed by automated tonality analysis tools.

To illustrate the analysis model, a representative article "Amendments to the Constitution of Yakutia have preserved all guarantees of free education" from the government source "Yakutia.Daily» ^[16].

Stage 1. General characteristics of the text. The text is journalistic, formed by the state media, the authorship is impersonal, collective editorial.

Stage 2. Establishment of semantic dominants. The object of discussion: amendments to the Constitution of the RS (Ya). The peripheral object is the education system.

Stage 3. Identification of tonality indicators. Mainly at the lexical level, tonality is manifested in such lexical units as "was accepted", "associated with absence", "explained", "does not guarantee", "allows", "does not cancel", "guaranteed", "why", "legitimize", "there is an opportunity", "also", "has the right", "accepted".

Stage 4. Interpretation of the tonal strategy. In these examples, the tonality is largely smoothed out by the informative narrative of the article. Nevertheless, at different levels, it is possible to catch such, at first glance, negative elements with the particle "not", which in the general context are used to maintain a general approving tone from the point of view of evidence. The tone is reinforced by positively labeled units such as "allows", "guarantees", "legitimize". An emotion of confidence and approval is formed. The polarity is moderately positive.

Bottom line: The main tone is moderately positive, approving, formally businesslike, informative, objectively collective, and unison. Peripheral or additional – absent. The goal is to set up readers positively and create an approving response.

Stage 5. Comparison with automated tonality analysis systems.

The Sentistrength system is based on marked-up tonal dictionaries. The system has not only the function of determining the polarity, but is also able to set its intensity according to tonal dictionaries. The similarity with expert analysis is incomplete, the levels of analysis coincide only in terms of polarity and intensity. The key was marked on a scale (-1; +1), and a moderately positive key of +1 was established.

Eureka Engine is shareware, and unlike specialized commercial media monitoring systems, it allows you to use a demo version. The functionality includes highlighting the objects of discussion, highlighting and labeling lexical polarity indicators, scaling from 0 to 1, where: 0-0.5 is a negative key, 0.5-1 is a positive key. The following subjects of discussion were highlighted in the text of the article: neutral tone – "Chapter", "profession", "amendment"; negative – "children", "education", "restriction"; positive – "amendment", "duty", "education". The general characteristic of tonality is positive, with an indicator of 0.78. The indicator of negative tonality is 0.08. This tool is closer to expert analysis in terms of detail, it provides the search for objects of discussion, there are parameters of polarity and its intensity.

YesChat Sentiment Analysis Classifier. This service is a modification of generative neural networks with prepared prompta for the end user. To solve the research questions, it is possible to set all the necessary analysis parameters, such as: definition of the communicative function, emotions, strategy of the speech genre; it is possible to obtain results in all parameters that repeat the expert analysis. For example, the system defines an expression with a positive tone "we keep all guarantees", identifies complex syntactic constructions, and the general emotional load of the article is characterized as "hope", "approval", the function of the article: informative, the formation of positive perception.

By contrast, the article "What are you doing with a circus here?", by Aisen Nikolaev, is given as a second example. Amendments to the Constitution of Yakutia have been adopted (Sakhaday.ru ) ^[17].

Stage 1. General characteristics. The text is a journalistic report published in a private independent media resource. The author takes a critical position, building the narrative within the framework of a conflict discourse, where the main focus is not on the very fact of the adoption of amendments, but on the political confrontation around it.

Stage 2. Establishment of semantic dominants. There are two poles at the center of the analysis: the head of the republic, Aisen Nikolaev (as the initiator and defender of the amendments), and his opponents (deputies, the public). Amendments, education, and asset management issues are the objects of discussion to demonstrate this split.

Stage 3. Identification of tonality indicators ^[12]. The tone is formed due to contrasting vocabulary: expressions conveying irritation are used in relation to the authorities ("it's funny to listen", "to reason with critics"). The vocabulary characterizing the position of the head of the republic is normative and well-reasoned: he appeals to the "federal law", the need to "bring legislation into line" and his constitutional duty as a guarantor. The rhetorical question "Why are you making a circus here?" is included in the title, although it is perceived as emotional. However, in the context of the entire statement, it serves as a reaction to criticism that is unconstructive, in the speaker's opinion, and emphasizes the seriousness and responsibility of the legislative process. Thus, linguistic means create a picture of a tense but working discussion, during which the official position is presented through the prism of legal argumentation. The description of the process itself ("the meeting was heated", "the discussion that arose") and the active citation create the effect of public tension and drama.

Stage 4. Interpretation of the tonal strategy. The final tone of the text is conflictual and dissonant. She is ambivalent: she sounds condescending and negative towards the authorities, and sympathetic towards her opponents. The author's goal is not to inform, but to accentuate the political split and question the legitimacy of the government's actions, involving the reader in polemics.

Stage 5. Comparison with automated systems.

• SentiStrength assigned the text an average positive score, identifying formal and procedural vocabulary ("accepted", "supported"), but completely ignoring contextually negative markers (for example, "staged a circus") and not recognizing the conflicting nature of the text.
• Eureka Engine assessed the overall tone as "neutral-positive", highlighting multidirectional clusters of vocabulary, but was unable to establish a connection between them and reconstruct a single bipolar evaluation strategy.
• YesChat Sentiment Analysis Classifier showed the most accurate result, determining the "mixed" tone and the presence of two evaluative focuses (power and opposition), as well as correctly indicating the emotions of irritation and criticism. However, this system also failed to fully interpret stylistic techniques (for example, the rhetorical question in the title) as tools of disapproval.

Thus, all systems successfully perform lexical segmentation and primary classification by polarity, but demonstrate a systemic inability to discursive-contextual interpretation. No algorithm was able to adequately recognize the key ambivalence and dissonant strategy of the text, reducing the analysis to simplified statistics.

For a comprehensive visualization of the patterns identified during the study, as well as to demonstrate the gap between superficial statistics and discursive reality, a summary table is presented below. It summarizes the results of quantitative (automated) and qualitative (expert) tonality analysis in a sample of 50 articles grouped by media type.

Table 2. Summary statistics on the tone of media coverage of the amendments to the Constitution of the RS(Ya)

Table 2 clearly demonstrates the discrepancies between the automatic tone assessment (row 1) and the strategies determined by expert analysis (row 2), which confirms the key hypothesis of the article. Automatic systems assign a positive rating to the majority of articles in state-owned media (89%) and interregional media (78%). However, expert analysis shows that these are fundamentally different positivities.: unison legitimization and rationalizing condescension. Without expert intervention, these strategies in the report will be mixed into one category of "positive tone". The automatic system correctly identified a cluster of negative articles (16%), which coincided with the expert classification of them as a conflict-exposing strategy. This confirms the usefulness of AI for primary screening and identification of "hot spots". The table shows that the topic of amendments is not monolithic. In the state media, the focus is shifted to education and the figure of the Head, while in the critical media equal attention is paid to local self-government, education and precious metals, and the image of the critical public is actively constructed (100%).

The last row of the table "Recommended stage of in-depth analysis" illustrates the proposed approach. Automation allows you to quickly process all 50 articles, identify clusters and basic trends. However, to understand the meaning of these trends, a full expert analysis is necessary for critical and interregional media articles (20 articles, 40% of the sample), where tonal strategies are the most complex and significant. For state-owned media and neutral private media, spot checks are sufficient to control the quality of automatic classification.

Conclusions

Thus, a comparison of the expert approach and automated tools allows us to identify common points that can serve to optimize labor costs in analyzing tonality. The expert approach is detailed according to a multifaceted analysis system and can be adapted to user requirements. In this particular case, indicators are highlighted in the text, their interrelation is built, in a contextual and discursive way. The media bias is also taken into account, which determines the presentation, tonality and unison in relation to a particular topic or subject of assessment. While automated systems identify evaluation objects, evaluation features, and indicators related to these areas group them. Then, according to the tonal dictionary, their tonality is determined on a scale of "negative-positive". Tasks such as determining variables, determining polarity, and statistical calculations can be performed by an automated system, which will speed up work when processing large text data, which is especially important for real-time batch analysis. The polarity groups are determined fairly accurately based on tonal dictionaries. What the system lacks is the possibility of a holistic, relief detailing of the overall picture, taking into account the re-accentuation of tonality caused by the modulation of context, figurative vocabulary and the peculiarities of discourse. Thus, it is proposed to distinguish two main stages: the stage of automated preprocessing and the stage of linguistic expert elaboration. The use of tonality analysis systems can optimize labor costs when processing big data, reducing time at the stage of selecting evaluation objects and tonality indicators. This allows you to further transfer the processed information for in-depth expert evaluation.

The present study contributes to an urgent interdisciplinary issue at the intersection of computational linguistics, media studies, and political analysis. The work demonstrated that in the context of the digitalization of the media environment and the growth of information volumes, the primacy of only quantitative, statistical methods of analyzing tonality is a methodological mistake, especially when working with a complexly structured regional discourse saturated with implicit assessments, cultural codes and political bias.

The article is published in the version approved by the reviewers (after receiving a positive review recommending the manuscript for publication) with corrections made by the author (after receiving the editor’s comments, if any).
Read all reviews on this article