Analysis of the mutual dynamics of stock quotes and the sentiment of textual mentions in the media of the company "OZON Holdings PLC" using correlation and sentiment analysis.

Shiboldenkov Vladimir Aleksandrovich PhD in Economics Associate Professor; Department of Business Engineering and Management; Bauman Moscow State Technical University 5, 2nd Baumanskaya str., building 1, Basmanny district, Moscow, 105005, Russia vshiboldenkov@bmstu.ru Tyurnev Aleksander Nikolaevich Student; Faculty of Engineering Business and Management; Bauman Moscow State Technical University (National Research University) 1206 Panfilovsky ave., 238 block, Zelenograd, Moscow, Russia alexander.tyurnev@gmail.com Afanasev Kirill Mironovich Student; Faculty of Engineering Business and Management; Bauman Moscow State Technical University (National Research University) 5, 2nd Baumanskaya str., building 1, Basmanny district, Moscow, 105005, Russia akirill2003@list.ru

Presnyakov Artem Olegovich Student; Faculty of Engineering Business and Management; Bauman Moscow State Technical University (National Research University) 5, 2nd Baumanskaya str., building 1, Basmanny district, Moscow, 105005, Russia presnart@yandex.ru

Introduction

In today's digital economy, brand image plays a key role in achieving a company's financial success. Consumers actively use online platforms to obtain information about products and services, as well as to form their own opinions about brands. In this regard, the analysis of the tone of mentions of the company in the media and social networks is becoming increasingly relevant for assessing the reputation and forecasting its financial results. However, despite the apparent link between the media image and financial performance, quantifying this impact, especially for fast-growing technology companies such as Ozon operating in a highly competitive environment and facing the need to continuously attract investment and retain customers, remains a challenge. The reason for this lies in the multifactorial pricing of stocks and the difficulty of distinguishing the influence of the media component. Existing approaches do not always allow for rapid response to changes in the information background, which, as a result, creates risks for investors and the company itself (for example, complicating long-term planning or increasing the cost of capital). This study aims to solve this problem by developing and testing a case-specific analysis methodology.

This article is devoted to the study of the relationship between the tone of mentions in the media and the financial performance of the company using the example of Ozon. Ozon, being one of the leading Russian online retailers, is actively present in the media space and faces constant attention from the public, which makes it particularly sensitive to fluctuations in public opinion. As a result, the company is forced to pay increased attention to managing its reputation, as a negative information background can lead to an outflow of customers to competitors or a decrease in the loyalty of sellers on the platform. Analyzing the relationship between brand perception in the media and financial performance will allow us to better understand how public opinion affects the financial stability and development prospects of the company. The relevance of such an analysis for a potential, albeit relatively narrow, readership, namely investors, financial analysts, PR specialists and the management of Ozon itself is determined by the need to obtain practically applicable decision-making tools. For example, understanding the impact of negative news can help Ozon allocate resources more effectively for crisis communications. On a broader, macroeconomic scale, the study is significant due to Ozon's key role in the Russian e-commerce market, and understanding the dynamics of its shares under the influence of media factors is important for assessing the health of the entire sector. A common problem for the e-commerce industry is its high sensitivity to consumer sentiment and the news background, and this study suggests one approach to quantifying it.

The purpose of this work is to analyze the influence of the tone of the company's mention in the media on the dynamics of stock prices of Ozon Holdings PLC. Using media data and stock quotes will allow for a comprehensive assessment, identify key factors affecting the financial stability of companies, and determine how brand perception affects their financial performance.

The research will use data from Ozon's financial statements, as well as publications about the company in the media. Data collection from the media will be carried out using web scraping and text analysis methods. Sentiment analysis methods based on machine learning algorithms, as well as statistical methods for correlation estimation, will be used to analyze tonality.

Ozon's activities in the dynamic e-commerce market are fraught with a number of existing problems, both specific to the company and common to the industry, including high stock volatility, dependence on investor sentiment, difficulties in reputation management in a highly competitive environment, the need to attract investments for growth, and the challenges of scaling operations while maintaining quality. This study shows that media analysis, in particular the analysis of the tone of mentions, offers Ozon practically significant ways to solve these problems. It allows you to better understand the causes of price fluctuations, identify "pain points" for operational improvements, build a positive reputation for raising capital, and integrate sentiment analysis data into operational monitoring to prevent reputational crises. Thus, media analysis becomes a tool for more effective risk management, making informed strategic decisions and increasing the overall investment attractiveness of the company.

It is expected that the results of the study will allow not only to close the existing gap in understanding the specifics of the media impact on Ozon, but also to propose a methodology that can be adapted for other public companies. This makes the study in demand for specialists seeking to improve the accuracy of financial forecasts and the effectiveness of reputation management in a modern information environment. For Ozon, as an object of research and one of the market leaders, the findings can become the basis for adjusting the communication policy and strategy of interaction with investors. At the macro level, the demonstration of the influence of media tonality on the capitalization of such a large player underscores the importance of information openness and public opinion management for the sustainable development of Russian business. Also, the results of the study will allow:

- identify the key factors influencing the tone of Ozon mentions in the media;

- determine how the tone of the media correlates with the financial performance of the company;

- draw conclusions about the impact of the brand's image on the financial stability and development prospects of Ozon;

- develop recommendations for the company on reputation management and improving the effectiveness of public relations.

The practical significance of the research lies in the possibility of using the results obtained to develop a company's reputation management strategy, optimize communication policy and increase its financial stability. This importance is especially high for the target audience interested in specific assessment and forecasting tools, as well as for understanding more general trends in the impact of the information field on business in Russia.

It is important to note that this study focuses on analyzing the relationship between media tonality and financial performance, without claiming to establish a direct causal relationship between these factors.

Literature review and research hypotheses

The impact of brand image on a company's financial performance is a complex and multifaceted issue that attracts the attention of researchers from various fields, including marketing, finance, sociology, and data science. Numerous studies confirm that a strong brand image can lead to increased revenue, increased customer loyalty, and stronger market positions, and media and social media play an important role in shaping public opinion and perception of brands, having a potential impact on investment decisions and stock prices, allowing qualified investors and market experts to use celebrity posts in social networks, notes and articles in popular media to make a profit.

One of the key aspects of brand image research is the analysis of the tonality of mentions in mass media (mass media) and social networks. Sentiment analysis, based on natural language processing (NLP) and machine learning methods, allows you to automatically determine the emotional coloring of texts and classify them as positive, negative or neutral.

In studies devoted to the analysis of the impact of textual information on economic indicators, special attention is paid to the use of Data Mining and Sentiment analysis methods. In particular, Mikhnenko's work shows how Data Mining tools can be used to analyze textual information from companies' annual reports and identify latent changes in their strategic direction. The study highlights the importance of analyzing the dynamics of key lexemes and their contexts, which serve as indicators of a non-numeric nature and reflect the deep processes taking place in companies. Such an analysis makes it possible to identify transformations in strategic orientations, as well as long-term trends in the development of companies, providing an opportunity to go beyond quantitative indicators and gain a deeper understanding of strategic changes. Examples are the dynamics of the decline of the lexeme "bank" and the dynamic growth of the lexeme "Savings Bank" in the reports of PJSC Sberbank, as well as the increase in the frequency of the acronym GRI in the reports of PJSC Inter RAO, which indicates rebranding and trends in declaring the principles of the Global Reporting Initiative, respectively. These results highlight the importance of text analysis for understanding long-term trends in the company's development. Although Mikhnenko's work demonstrates the value of text analysis for understanding strategic shifts, it focuses on annual reports that reflect fait accompli with a certain time lag. Our research, on the contrary, is aimed at analyzing operational media mentions that can have a faster impact on stock prices, which is especially important for the dynamic e-commerce market ^[1].

Bogdanov and Dulya have developed an innovative approach to sentiment analysis of short Russian-language texts in social media, which is based on vector representation of text and the use of machine learning methods. This study demonstrates the effectiveness of sentiment analysis for marketing research and monitoring audience loyalty, which is directly related to our task of analyzing the impact of the tone of mentions in the media on the dynamics of stock prices. They have shown that their approach provides an automated solution for analyzing large amounts of data from social media. Their research also demonstrates how vector representation of text combined with machine learning methods can be effectively used to automatically classify the emotional coloring of a text. Therefore, the work of Bogdanov and Dooley makes a significant contribution to the development of methods of sentiment analysis of Russian-language texts. Their approach is a valuable tool for marketing research, monitoring audience loyalty, and, most importantly in the context of our work, for analyzing public opinion and its potential impact on companies' financial performance ^[2]. Bogdanov and Dooley's work makes a significant contribution to the methodology of sentiment analysis of Russian-language texts, and our approach to tonality analysis is largely based on similar machine learning principles. However, their research focused on social media, while we are expanding the scope to news media, which may have a different type of influence and a different audience. In addition, we strive not only to classify the tonality, but also to directly relate it to the dynamics of quotations of a particular company, such as Ozon.

In addition, numerous studies confirm the importance of intangible assets, such as brand and reputation, for the financial stability of companies. Zhukova and Melikova's work investigated the impact of corporate social responsibility (CSR) on the relationship between brand value and a company's financial performance. The authors have shown that investments in CSR can have a positive indirect impact on return on assets and market capitalization, and this impact is enhanced for companies with a higher level of social responsibility. This means that the positive reputation associated with CSR increases the financial return from a strong brand. At the same time, the study demonstrated that CSR has no impact on the relationship between brand and return on equity. In addition, the hypothesis about the positive impact of the brand on return on assets has not been confirmed, which indicates the difficulty of measuring the brand. These findings emphasize that the perception of a company in society can indirectly influence its financial results. Therefore, analyzing the tone of mentions in the media, shaping the perception of the brand, can also have an impact on the company's share price. The work of Zhukova and Melikova emphasizes the importance of studying not only direct, but also indirect links between intangible assets, reputation and financial results, which is the purpose of our research ^[3]. The results of Zhukova and Melikova indicate an indirect influence of reputation on financial performance. Our study complements these findings by looking at a more direct channel of influence – the tone of media mentions - and focusing not on general profitability indicators, but on a specific market indicator – the stock price. The fact that the hypothesis of the brand's impact on return on assets has not been confirmed in their work also highlights the complexity of measurement and the need to find more sensitive indicators and methods, which is what we are trying to do.

The image of an organization is an important factor influencing its competitiveness and financial stability. For example, Likhonos noted in his work that a positive image of an organization facilitates access to resources and contributes to improving its financial performance. His research confirms that a favorable image increases the company's competitiveness in the market, attracts customers and suppliers, accelerates sales, increases their volume and, thus, positively affects financial results. Likhonos also pointed out that a strong image builds customer loyalty and provides higher income, reducing risks. These results emphasize the importance of image management and, in the context of our study, indicate that the tone of media mentions, shaping the perception of a company's image, can have an impact on its financial performance. Likhonos' research confirms the need to analyze not only financial indicators, but also the perception of the company in the media space in order to understand the factors affecting its financial well-being. Likhonos' conclusions about the importance of image for financial results are fully consistent with the premises of our research. However, his work is more general in nature, without delving into the specifics of quantifying the impact of the tone of media messages on the dynamics of stock prices of a public company in the Russian market, which is the focus of our article ^[4].

The study by Zorin and Sadkovkin highlights that publications in the media and social networks, including the opinions of stars and opinion leaders, can significantly influence stock prices. For example, Elon Musk's tweet about Tesla's collaboration with Panasonic led to a drop in Samsung SDI shares and a rise in Panasonic shares. Other examples include the jump in Tesla stock prices after Musk's announcement of a possible share buyback, the collapse of Freedom Holdings shares after the Hindenburg Research investigation, and the impact of Donald Trump's Twitter posts on the market. The media and social media form public opinion, which influences investment decisions, and qualified investors can use this to make a profit. Media publications can also influence the course of negotiations on mergers and acquisitions (M&A) transactions. The examples given by Zorin and Sadkovkina vividly illustrate the power of the influence of individual news events. Our study seeks to systematize this effect by moving from the analysis of individual cases to the identification of statistically significant patterns based on the analysis of a large array of media data over a long period for a particular Ozon company, which will allow us to assess not only the event, but also the background influence of media tonality ^[5].

A study by Fedorova, Lapshina, Lazarev, and Borodin demonstrated that the tone of press releases has an impact on the financial performance of Russian companies. In particular, they found that the positive tone of press releases can lead to a positive reaction from investors and an increase in stock prices. Their work confirms that investors trust press releases about future results and tend to respond positively to the positive tone of publications, as well as to statements related to the disclosure of business prospects. In addition, the study showed that press releases are becoming more detailed, and they use words describing the future of the company, which has a positive effect on the effectiveness of the company. This indicates that the tone of a company's publications can influence investors' perception of its prospects and, consequently, the share price. In this regard, the results of Fedorova, Lapshina, Lazarev and Borodin confirm the need to analyze not only financial indicators, but also the linguistic characteristics of publications, and also indicate the importance of studying the impact of public opinion on the financial well-being of the company, which is the purpose of our research. The work of Fedorova et al. is close to ours in terms of subject matter, but their focus is on official press releases of companies. We analyze a wider range of media mentions, including independent news articles that may reflect not only the company's position, but also a critical assessment of its activities, which, in our opinion, is a more complete reflection of public opinion. Thus, our research expands the understanding of the impact of textual information, going beyond the company-controlled messages ^[6].

In particular, S.A. Romanov's work examines the impact of media mentions on the financial performance of the banking sector. In this study, the main focus is on the number of mentions of the bank in the media, rather than on the tone of these mentions, and it is argued that the very fact of the frequent appearance of the bank's name in the media has a positive effect on its operating income. Correlation analysis showed a direct linear relationship between the number of mentions of Sberbank PJSC in the Russian-language media and its operating income. The author of the paper suggests that frequent mention of the bank in the media increases its recognition and interest in its offers, even if the information occasion is neutral or moderately negative, but does not take into account the impact of sharply negative information. The study highlights that the number of media mentions alone can be an important factor influencing a bank's financial results. This approach complements other studies that examine the impact of the tone of media publications on financial results, showing that not only the perception of a brand through the prism of positive or negative assessments, but also the level of media attention itself can significantly affect the economic performance of a company. Romanov S.A.'s work is interesting because it focuses on the quantitative aspect of media-presence. In contrast, our research focuses on a qualitative characteristic – tonality. We believe that for technology companies like Ozon, where innovation and reputation play a key role, not just the frequency of mentions, but their emotional coloring is a stronger driver of investor sentiment and, as a result, stock prices. Thus, we strive to show that taking into account tonality gives an additional increase in understanding market mechanisms compared to analyzing only the number of publications ^[7].

The following review examines 23 scientific articles on stock price forecasting and the impact of investor sentiment on the stock market, with an emphasis on their practical significance and the tasks to be solved. This block of research ^[8-17] confirms the global interest in forecasting markets using sentiment data extracted from various sources, including news articles and social media. Many of these works (for example, ^{[9],[10],[15]}) NLP and machine learning methods are successfully applied, which corresponds to our methodological approach. However, most of them are either focused on Western markets, or use specific sentiment indices (for example, from Twitter ^[15],[17]), or analyze other asset classes (oil ^[8], bitcoin ^[11]). Our research contributes by applying modern NLP models to Russian-language news texts to analyze the shares of a particular Russian e-commerce company, which is a less studied area.

Oil price forecasting ^[8] emphasizes the importance of evaluating oil price forecasting models not only in terms of statistical accuracy, but also in terms of their practical usefulness for making financial decisions. It is proposed to use economic indicators such as equivalent profitability to evaluate the effectiveness of oil price forecasting models, which will allow investors to make more informed decisions.

The effect of word meaning disambiguation on stock price forecasting ^[9] demonstrates that taking into account the meaning of ambiguous words in news articles using WSD can improve the accuracy of stock price forecasting. It is proposed to use the NLP pipeline based on WSD to analyze financial news and predict its impact on stock prices.

The use of news broadcasts to predict stock prices of companies operating in the field of renewable energy sources ^[10] shows that data from television news broadcasts about the environment can be used to predict stock prices of companies operating in the field of renewable energy sources. It is proposed to use the sentiment analysis of news broadcasts as input data for forecasting models based on LSTM in order to improve the accuracy of forecasting.

The impact of investor sentiment on risk in the Bitcoin market ^[11] highlights the importance of taking investor sentiment into account when assessing risks in the bitcoin market. It is proposed to use ChatGPT to analyze sentiment in news articles and apply KNN regression to build a risk forecasting model in the bitcoin market.

The role of sentiment analysis in brand management and marketing ^[12] demonstrates how sentiment analysis of customer reviews can be used to improve brand management, marketing strategies, and increase customer satisfaction. It is proposed to use various sentiment classification methods such as SVM, naive Bayes method, RNN and LSTM to analyze customer reviews and identify their attitudes towards products and services.

The impact of retail investors' search for investment ideas on social media on stock prices ^[13] reveals the impact of retail investors' search for investment ideas on social media on stock price dynamics, which is important for both retail and institutional investors to consider. It is proposed to use sentiment analysis and the context of tweets, as well as various machine learning methods, to predict stock prices, trading volume and volatility.

The use of transfer entropy to determine the causal relationship between news sentiment and stock prices ^[14] demonstrates the advantages of nonparametric methods such as Shannon and Renyi transfer entropy for analyzing causal relationships between financial news sentiment and stock prices. It is proposed to use transfer entropy to more accurately determine the effect of sentiment on market movement compared to traditional methods such as the Granger test.

The use of LSTM to predict stock prices based on sentiment ^[15] demonstrates the potential of LSTM to predict stock prices based on investor sentiment. It is proposed to use sentiment data from Twitter and optimize the LSTM architecture to achieve the best valuation performance when predicting stock prices.

The influence of microblogging sentiment on stock price dynamics ^[16] shows that sentiment expressed in microblogs can have a significant impact on stock price dynamics. It is proposed to use microblogging sentiment analysis and social media analysis methods to study the relationship between sentiment and price changes, which can be useful for making investment decisions.

Forecasting stock prices using the sentiment index ^[17] demonstrates the potential of sentiment indices based on Twitter and Google Trends data to predict stock prices. It is proposed to use machine learning and deep learning methods to build price forecasting models based on sentiment indices, which can improve the accuracy of forecasts compared to traditional time series methods.

The article ^[19] provides valuable information on the methods of analyzing the tonality of financial texts that can be used to assess the tonality of Ozon mentions in the media. The article also highlights the importance of taking into account contextual information, which is especially important when analyzing stock price dynamics. The practical significance lies in the development of a new methodology for fine-grained analysis of the tonality of financial texts, which takes into account contextual information. The article solves the following practical tasks: improving the accuracy of tonality analysis by using information about the relationships between financial texts, as well as the semantics and linguistic features of the texts.

The article ^[26] is devoted to the study of the impact of extended audit reports, in particular the section "Key Audit issues" (KAM), on investor sentiment and reaction in Thailand. Practical tasks: to identify the relationship between the characteristics of KAM (quantity, type, tonality) and the reaction of investors in the stock market. The article demonstrates a methodology for analyzing the tonality of texts developed specifically for the Thai language. The described approach using a "basket of words" and a survey of experts can be adapted to analyze the tonality of Russian-language texts about Ozon.

The authors of ^[30] show a practical example of analyzing the relationship between investor sentiment and stock price dynamics. The methods of data collection (using an Internet crawler) and tonality analysis (using neural networks) can also be used for Ozon research. The article is devoted to the development of a stock trend forecasting model using investor sentiment analysis extracted from news data. Practical tasks: to create a model that can take into account the impact of news on the dynamics of stock prices.

The following article ^[29]. It focuses on analyzing the tone of forum posts, which can be useful for studying investor sentiment on online platforms such as forums or social networks. The authors devoted the article to studying the impact of user content on investor forums on the GEM index. To do this, they solved the following tasks: developing a machine learning model for analyzing the tone of forum posts and its relationship to the dynamics of the index.

The study ^[22] on the influence of investor sentiment on the innovative activity of companies determines the causal relationship between investor sentiment and the number of patents and citations using text analysis. The article demonstrates the use of econometric methods to analyze the relationship between investor sentiment derived from text data and company performance indicators.

The article ^[21] presents a bibliometric analysis of studies that study the impact of online investor sentiment on financial markets. This paper identifies the main topics, authors, journals, and methods in the field, as well as highlights knowledge gaps and directions for future research.

The work ^[20] analyzes how user content on social networks affects stock price forecasts, using GameStop as an example. The study developed a methodology for integrating data from social networks with financial data to create a predictive model that uses machine learning methods. This approach can be used to assess the impact of the tone of Ozon mentions on social media on the dynamics of its stock prices.

The study ^[27] analyzes the tone of Twitter messages to assess financial stability in Mexico. The authors have created a risk monitoring system based on data from social networks. This method can be applied to analyze Ozon's financial performance using data from social media.^[4]

The article ^[25] discusses the use of graph neural networks (GNNs) to predict stock prices.4 The authors compare the advantages of GNN with traditional methods, as well as consider the challenges and future research directions. This study provides information on the use of GNN to study the relationship between Ozon stock quotes and the tone of media mentions.

The article ^[23] is devoted to the development of a method for analyzing the tonality of financial texts using vocabulary and hints. The main purpose of this work is to increase the accuracy of tonality analysis by taking into account the specifics of financial vocabulary. The proposed method can be adapted to study texts about Ozon.

Taken together, these two articles suggest methods that can be used for a deeper analysis of the impact of both the emotional background on social media and the tone of financial texts on the dynamics of Ozon shares.

In the article "Improving sentiment analysis of financial news headlines using hybrid Word2Vec-TFIDF feature extraction technique" ^[24], the authors explore methods to improve the analysis of the tonality of financial news headlines using the hybrid feature extraction technique Word2Vec-TFIDF.

The practical significance of the study is due to the fact that the analysis of the tone of financial news plays an important role in making investment decisions and managing risks. A more accurate analysis allows you to:

- predict stock price movements,

- assess investment risks,

- understand the market sentiment.

The authors used a dataset of 200501 financial news headlines from the Kaggle platform. The data was preprocessed using various methods such as lowercase conversion, removal of stop words and punctuation, stemming and lemmatization.

The article tested five feature extraction methods: Word2Vec, Doc2Vec, TFIDF, Doc2Vec-TFIDF and Word2Vec-TFIDF.

Six machine learning algorithms were used to classify tonality: random forest (RF), support vector machine (SVM), decision tree (DT), Gradient boosting (GB), Extra Trees (ET) and logistic regression (LR).

The results of the study showed that the hybrid Word2Vec-TFIDF method in combination with the SVM classifier provides the best accuracy (82%) in analyzing the tonality of financial news headlines.

The article "Beyond Negative and Positive: Exploring the Effects of Emotions in Social Media during the Stock Market Crash" ^[28] explores the impact of emotions expressed on social media on the stock market during the crash.

The practical significance of this research lies in the fact that it helps to understand how emotions in social networks can influence investor behavior and, consequently, market dynamics. This knowledge can be used for:

1. Better risk management: Understanding how emotions affect the market allows investors and financial institutions to develop strategies to mitigate the risks associated with emotional reactions to market events.

2. Developing trading strategies: Traders can use information about social media sentiment to make more informed investment decisions.

3. Forecasting market trends: Analyzing the emotional coloring of social media posts can help predict future market movements.

The authors collected more than 300,000 messages from the social network Weibo related to 34 companies whose shares are listed on the stock exchange. The emotional coloring of messages was assessed using a two-dimensional model of emotions that takes into account valence (positive-negative) and arousal (calmness-arousal). The authors used hidden Markov models (HMM) to determine the state of market perception (for example, a state of panic, a state of recovery).

Using an orderly logistic regression, a link was established between the emotions expressed on Weibo and the state of market perception.

The article emphasizes that the analysis of tonality, which goes beyond the simple definition of positive and negative assessments, is necessary to understand the impact of social networks on financial markets, especially during crises. Using a two-dimensional model of emotions and HMM provides new tools for analyzing the impact of social media on market perception.

The article "Stock Prediction by Integrating Sentiment Scores of Financial News and MLP-Regressor: A Machine Learning Approach" ^[18] is devoted to stock price prediction by integrating the analysis of financial news tonality and a multilayer perceptron-regressor. The practical significance of this work lies in the fact that it provides a new tool for investors and traders that can help them make more informed investment decisions. A more accurate prediction of stock price movements can lead to increased profits. Understanding the impact of financial news on the market allows investors to keep abreast of events and respond to them more quickly.

The authors used 10 years of historical stock price data and financial news from four companies from different sectors. Three algorithms were used to calculate tonality ratings: VADER, TextBlob, and Flair. The sentiment estimates were combined with historical stock price data.

A multi-layered perceptron regressor was trained on a combined dataset to predict future stock prices. Various metrics such as accuracy and average absolute percentage error (MAPE) were used to evaluate the effectiveness of the model.

The article shows that the integration of financial news tone analysis and machine learning methods can significantly improve the accuracy of stock price prediction.

In general, the article "Stock Prediction by Integrating Sentiment Scores of Financial News and MLP-Regressor: A Machine Learning Approach" is a valuable contribution to the research topic, as it demonstrates the potential of using financial news tone analysis to improve the accuracy of stock price forecasting models.

The following conclusions can be drawn from this. Investor sentiment, extracted from social media and news sources, is an important factor influencing stock prices. Modern machine learning and deep learning methods are effective in predicting stock prices based on sentiment.

Nonparametric causal analysis methods, such as transfer entropy, represent a promising direction for studying the relationship between sentiment and financial performance.

The authors identify the following practical recommendations. Investors should take into account the sentiments expressed on social media and news sources when making investment decisions. Analysts and researchers should use modern machine learning and deep learning techniques to build stock price forecasting models. It is important to carefully select and optimize the hyperparameters of the models to achieve high prediction accuracy.

Thus, an analysis of the existing literature shows that, despite the recognition of the importance of media influence and tonality on financial markets, there are still a number of open questions and insufficiently studied aspects. Firstly, most of the research focuses on developed Western markets, while the specifics of the Russian information field and its impact on domestic companies require additional study. Secondly, although there are studies on the analysis of the tonality of Russian-language texts, their direct relationship to the dynamics of stock prices of specific companies, especially in the e-commerce sector, remains insufficiently investigated. Thirdly, many studies either analyze individual information events or use general sentiment indexes, whereas our study aims to identify a systematic relationship between the aggregated quarterly tonality of a wide range of media and Ozon's financial performance. In addition, the DeepPavlov/rubert-base-cased-conversational model we use for sentiment analysis of Russian-language texts is more modern than some of the approaches described in earlier works.

Based on the literature review and analysis of available data, we propose the following hypothesis:

Hypothesis: there is a statistically significant correlation between the dynamics of Ozon Holdings PLC stock quotes and the tone of text mentions of the company in the media. It is expected that a positive tone of mentions in the media will be associated with an increase in stock prices, while a negative tone will be associated with a fall in stock prices.

The market value of shares of a public company such as Ozon is influenced by many factors, including financial performance, the general economic situation, the actions of competitors and, most importantly, the public perception of the company formed in the media space. Negative news and media sentiment can quickly undermine investor confidence, lead to a drop in stock prices and make it difficult to raise capital. Conversely, positive coverage helps to increase confidence, increase demand for shares and strengthen the company's position in the market.

Why is this a problem for Ozon:

High stock volatility: Ozon, as a fast-growing technology company, is particularly sensitive to changes in investor sentiment and public opinion. Unexpected jumps and drops in quotations can complicate long-term planning, reduce the attractiveness for investors and complicate the assessment of the real value of the company.

Reputation management: negative publications in the media can damage the brand's reputation, scare off customers and partners, which will eventually negatively affect financial results.

Attracting investments: investors make decisions based not only on financial statements, but also on the overall perception of the company. A negative information background may alienate potential investors or require the company to raise higher capital costs.

Lack of operational information: traditional methods of financial market analysis do not always allow us to quickly monitor and evaluate the impact of the information field on stock prices. Information spreads quickly, especially in a digital environment, and companies need to have the tools to assess and respond quickly.

Thus, the conducted tonality analysis helps Ozon Holdings PLC and other public companies.:

• Better understand the factors influencing stock volatility;

• manage your reputation in the information space more effectively;

• make more informed decisions in the field of communications and investor relations;

• reduce the risks associated with a negative information background;

• Improve the investment climate and make it more attractive to investors.

Research methodology

To achieve this goal, the following tasks were defined:

• Collection and preparation of quotation data: collection and preparation of quarterly data on share prices of Ozon Holdings PLC for the period from 2021 to 2024 (inclusive). The data will include the opening, high, low, and closing prices, as well as the turnover (trading volume) for each quarter. The Moscow Stock Exchange will serve as the data source.

• Collection and preparation of text data: collection and preparation of a corpus of text data from the media (Google News) containing mentions of Ozon Holdings PLC for the period from 2021 to 2024 (inclusive). The text data will be preprocessed for further analysis.

• Tonality analysis: analyzing the tonality of the collected text data using the DeepPavlov/rubert-base-cased-conversational model. The result of the analysis will be the indicators "Dry tonality" and "Dry tonality Adjusted", calculated on the basis of normalized values of "Positive Normalized", "Negative Normalized" and "Neutral Normalized" for each quarter, where the normalized value reflects the percentage of the criterion to the total number of mentions.

• Correlation Analysis: Performing correlation analysis to assess the relationship between stock price changes ("Dynamic OC") and sentiment estimates ("Dry Key" and "Adjusted Dry Key") for each quarter. The correlation between tonality indicators and other financial indicators of stocks such as "Open", "High", "Low", "Close", "Avg" (average) and "Turnover" will also be analyzed.

Let's consider the approach to calculating the necessary indicators in more detail. Formulas:

Positive Normalized = (Number of positive mentions / Total number of mentions) * 100% (1.1)

Negative Normalized = (Number of negative mentions / Total number of mentions) * 100% (1.2)

Neutral Normalized = (Number of neutral mentions / Total number of mentions) * 100% (1.3)

Dry Tonality = ((Positive Normalized - Negative Normalized) / Neutral Normalized) * 100% (1.4)

Dynamic OC = Close Price - Open Price (1.5)

Dry key (adjusted)= Dry key + C (1.6)

where C is the constant for the Y–axis offset.

Next, we will construct a DFD scheme (see Figure 1).

Figure 1 – DFD diagram of the research methodology

The diagram provided shows a diagram of the research methodology, probably devoted to the analysis of the tonality of text data and its relationship to the dynamics of stock prices, in particular, OZON Holding PLC. The diagram shows the sequence of data processing steps, starting with parsing Google search results and ending with correlation analysis.

This study uses a multi-stage approach to data analysis, which includes the following steps::

• Data collection from the Google search engine: at the first stage, Google search results are parsed using a specialized script. The data is collected for 16 quarters (from 2021 to 2024) and is a table of links to web pages.

• Web page parsing: The resulting links are used to parse the contents of the corresponding web pages. The result of this stage is an aggregated array of texts of articles.

• Data filtering: the resulting array of texts is double filtered:

• Date filtering (by quarters from 2021 to 2024).

• Filtering by language (selection of Russian-language articles).

• Definition of tonality: for each text, its tonality is determined. The results are saved in a CSV file format with details on the tone in the context of the articles.

• Tonality aggregation: based on the data obtained, the total tonality is calculated by quarter. MS Excel is used to process and analyze this data.

• Tonality analysis (Dry Tonality): The values of Dry tonality are calculated, which is probably a specific method of tonality analysis used in this study.

• Stock price data collection: in parallel with the analysis of text data, quarterly indicators of OZON Holding PLC securities from the Moscow Stock Exchange are collected.

• Correlation analysis: At the final stage, a correlation analysis is performed between the price dynamics of OZON Holding PLC shares and the calculated tonality values (including Dry Tonality). The results of the analysis allow us to establish the presence or absence of a correlation between these two variables.

The results of the study

Table 1 presents aggregated data on quotations of shares of Ozon Holdings PLC obtained from the Moscow Stock Exchange. Including quarterly opening, maximum, minimum, closing and turnover prices for the period 2021-2024, the data from this table will be used to calculate the dynamics of price changes and to conduct further correlation analysis with media sentiment indicators.

The analysis of the dynamics of Ozon Holdings PLC stock quotations for the period from 2021 to 2024, presented in Table 1, demonstrates significant volatility and a downward trend in the long term. The peak of the share price occurred in the 1st quarter of 2021, when the opening price was 3269.5 rubles, and the maximum price for the quarter reached 5181 rubles.

However, already in the second half of 2021, a sharp decline in quotations began. In the 3rd quarter, the closing price was 3675.0 rubles, and by the end of the year, the shares were already trading at 2316.5 rubles. The decline continued in 2022, reaching a minimum of 728 rubles in the 2nd quarter.

In 2023, there is an increase in the share price. As of 31.12.2023, the share price of Ozon Holdings PLC averaged 2,804.5 rubles. In general, 2024 also ended with an increase in shares, where the price per share was 3,301 at the close of trading.

To analyze the tone of mentions of Ozon Holdings PLC in the media, text data was collected using the Google News search engine. To ensure relevance and relevance to the time period of the study, the search was carried out using the following criteria:

• Time filter: The search results were limited to the quarterly periods for 2021-2024 to match the structure of the financial data.

• Language filter: A filter was used that restricts search results to Russian-language articles to ensure compliance with the DeepPavlov/rubert-base-cased-conversational model.

To automate the process of collecting links to articles that meet the specified criteria, a simple JavaScript script was developed and used. It automatically extracted links to all articles that met the search criteria and saved them in a structured form.

As a result of applying these criteria and using an automated script, about 1,600 articles were received. A table with articles was created for each quarter. After that, the results were saved as a table.

Next, the collected text data was preprocessed to prepare for the tonality analysis. This stage included the following steps:

• Data cleanup: Rules have been developed and applied to delete strings containing junk data (for example, special characters, numbers, remnants of HTML markup) to improve the accuracy of analysis.

• Tonality assessment: After cleaning, the text data was transferred to the DeepPavlov/rubert-base-cased-conversational model to determine the tonality of each sentence. For each word, the mention was assigned to one of three categories: positive, neutral, or negative. For each word, the number of mentions in each key was calculated.

• Data aggregation: For further analysis, all data has been compiled into a single table containing quarterly results for each word.

• Final cleanup: After combining all the data into one table, all rows that are clearly not words, but represent residual garbage data, were deleted.

• Final aggregation: After cleaning, the data was grouped and summarized by quarter, and the number of words for each category was counted to obtain a summary table with the results. As a result, a table was obtained with the final quarterly values of the frequency, number of positive, negative and neutral words. The structure of the final table is presented in the form of an image.

The results are presented in table 2.

Since the number of mentions differs quantitatively in each quarter, the data needs to be normalized. To do this, use the normalization formulas 1.1, 1.2, 1.3. The results are shown in Table 3.

To obtain a single value reflecting the overall ratio of positive and negative in each quarter, the "Dry tone" indicator was calculated using formula 2.1. This indicator allows us to take into account not only the absolute number of positive and negative mentions, but also their ratio to neutral mentions. To calculate the Dry Key (adjusted) according to formula 2.2, we take the constant C as 0.05. The results are shown in Table 4.

A "dry tonality" indicates the net positivity of the mood, taking into account the neutrality, providing an assessment of the polarity of the mood relative to the neutral baseline. A negative value (-5.65%) indicates a slightly negative mood bias in general.

For clarity, let's reflect the data obtained on the graph (see Figure 2).

Figure 2 – Comparison of the dynamics of the tonality and the share price

It can be seen that the trend of the tonality coincides with the trend of the stock price, we calculate the correlation and statistical significance (see Table 5).

The high correlation index (72%) and statistical significance (0.07%) not only confirm the existence of a mathematical relationship between the dynamics of tonality and the dynamics of the company's shares, but also indicate a significant economic relationship. This suggests that changes in the public perception of Ozon, reflected in the tone of the media, can translate into measurable changes in the company's market capitalization. For example, an improvement in the “Dry Tone” indicator can potentially lead to an increase in the share price, creating economic benefits for shareholders, while its deterioration can lead to direct financial losses.

Conclusions

A study of the relationship between the tone of text mentions in the media and the dynamics of stock prices of Ozon Holdings PLC has revealed a number of important patterns. An analysis of the data collected from 2021 to 2024 showed a statistically significant correlation between the studied variables, which confirms Hypothesis 1.

The tone of media mentions has an impact on the dynamics of the company's stock prices. A positive tone reflecting a favorable perception of the company in the information space is associated with an increase in stock prices. A negative tone, signaling the presence of problems, can lead to a decrease in the value of shares.

Using the methods of sentiment analysis and correlation analysis allows us to identify this relationship and assess its strength. The use of the DeepPavlov/rubert-base-cased-conversational model for analyzing the tonality of texts in combination with the analysis of quarterly stock price data ensured high accuracy and statistical significance of the results obtained.

The revealed relationship has practical significance for the development of company reputation management strategies and optimization of communication policy. Monitoring and analyzing the tone of the media make it possible to respond in a timely manner to changes in the perception of the company by the public, adjust communication strategies and minimize the risks associated with a negative information background.

The results of the study are consistent with the conclusions of other authors who have studied the impact of the company's image and media tone on its financial performance.

Understanding the relationship between the tone of the media and the dynamics of Ozon Holdings PLC stock prices can be used to make more informed investment decisions. For Ozon Holdings PLC, the research results can be used to develop an effective reputation management strategy, proactively interact with the media, and create a positive brand image.

The study confirms the effectiveness of modern methods of text data analysis (sentiment analysis) for solving practical problems in the field of finance and reputation management.