Differential Impacts of Environmental, Social, and Governance News Sentiment on Corporate Financial Performance in the Global Market: An Analysis of Dynamic Industries Using Advanced Natural Language Processing Models ^†

Abstract

This study examines how sentiment analysis of environmental, social, and governance (ESG) news affects the financial performance of companies in innovative sectors such as mobility, technology, and renewable energy. Using approximately 9828 general ESG articles from Google News and approximately 140,000 company-specific ESG articles, we performed term frequency-inverse document frequency (TF-IDF) analysis to identify key ESG-related terms and visualize their materiality across industries. We then applied models such as bidirectional encoder representations from transformers (BERT), the robustly optimized BERT pretraining approach (RoBERTa), and big bidirectional encoder representations from transformers (BigBird) for multiclass sentiment analysis, and distilled BERT (DistilBERT), a lite BERT (ALBERT), tiny BERT (TinyBERT), and efficiently learning an encoder that classifies token replacements accurately (ELECTRA) for positive and negative sentiment identification. Sentiment analysis results were correlated with profitability, cash flow, and stability indicators over a three-year period (2019–2021). ESG ratings from Morgan Stanley Capital International (MSCI), a prominent provider that evaluates companies’ sustainability practices, further enriched our analysis. The results suggest that sentiment impacts financial performance differently across industries; for example, positive sentiment correlates with financial success in mobility and renewable energy, while consumer goods often show positive sentiment even with low environmental ESG scores. The study highlights the need for industry-specific ESG strategies, especially in dynamic sectors, and suggests future research directions to improve the accuracy of ESG sentiment analysis.

1. Introduction

The establishment of global sustainability standards and the 2021 Glasgow Climate Pact at the 26th United Nations Climate Change Conference of the Parties (COP26) underscored the importance of environmental, social, and governance (ESG) management for companies worldwide, particularly in sectors such as mobility, technology, and renewable energy [1]. These agreements emphasize the need for companies to actively pursue sustainable value management, aligning environmental and social responsibilities with carbon neutrality goals [2]. The introduction of these standards encourages companies to adopt transparent and responsible management practices, which are expected to significantly impact capital market disclosures and the investment environment, especially in dynamic industries such as the mobility sector.

Reflecting this global trend, South Korea has introduced new requirements for companies listed on the Korea Composite Stock Price Index (KOSPI): those with assets exceeding KRW 2 trillion (South Korean won) will be required to publish sustainability management reports starting in 2025, and this mandate will be extended to all KOSPI-listed companies by 2030 [3]. Although there are currently no legal penalties equivalent to financial reporting standards, these developments highlight the growing importance of ESG management, which requires systematic preparations by companies. However, the lack of clearly defined ESG standards has created confusion among companies and investors, with inconsistencies in the evaluation criteria leading to varying ESG ratings and increasing investor uncertainty [4].

Among various industries, ESG management is particularly critical in the mobility industry due to its significant environmental impacts [5]. With the emergence of innovative technologies such as electric vehicles, autonomous vehicles, and shared mobility services, companies in the sector need to develop targeted ESG strategies that directly address key challenges [6]. From an environmental perspective, companies should prioritize reducing carbon footprints and improving waste management to meet increasing regulatory and consumer demands for sustainable practices. Social responsibility strategies can include fair labor practices and increased community involvement. In addition, strengthening governance through transparent practices, such as implementing regular ESG reporting and establishing board diversity policies, is essential to meet growing stakeholder expectations for accountability [7]. Effective ESG management is inextricably linked to the long-term competitiveness of the mobility sector. To achieve this, it is essential to develop strategies that take into account key ESG drivers such as economic development, the regulatory environment, and responsible investment. Identifying these factors is critical to improving overall ESG practices [8].

This study presents a novel, comprehensive, cross-industry approach to ESG sentiment analysis using multiple natural language processing (NLP) models, building on existing research to explore an innovative method not yet fully developed [9]. By examining ESG sentiment across different sectors, each with unique ESG impacts and challenges, this study addresses a significant gap in the literature, as existing studies typically focus on a single industry or rely on a single-model approach to sentiment analysis [10,11]. In doing so, this study aims to analyze global ESG management trends and investigate the correlation between financial performance and ESG-related news sentiment across various industries, including the mobility, technology, and renewable energy sectors.

To achieve this, we compare and analyze sentiment models such as bidirectional encoder representations from transformers (BERT), decoding-enhanced BERT with disentangled attention (DeBERTa), financial BERT (FinBERT), and RoBERTa, using state-of-the-art NLP techniques to assess correlations with corporate financial indicators [12]. Specifically, we collected, verified, and evaluated data for companies in these sectors from approximately 700 companies listed on the New York Stock Exchange (NYSE), the Nasdaq Stock Market (NASDAQ), and the American Stock Exchange (AMEX) that have received ESG ratings from MSCI. This analysis enables a systematic examination of the impact of ESG-related sentiment on financial performance [13].

Building on previous research [9], the current study extends the analysis by integrating nine advanced NLP and sentiment analysis models along with a comprehensive set of financial indicators across multiple industries. Unlike previous studies, which often rely on limited models or specific financial metrics, this research uses a multimodel, multifaceted approach to provide a more comprehensive view of the impact of ESG sentiment on financial performance [10,11]. This approach not only increases the robustness of our findings but also addresses critical limitations of previous studies by providing a nuanced framework for evaluating ESG strategies across sectors.

Specifically, this study addresses the following research questions:

• How does positive or negative ESG-related news sentiment affect a company’s financial performance across different industries, especially in the mobility, technology, and renewable energy sectors?
• What specific patterns emerge in each industry regarding the impact of ESG sentiment on financial outcomes, and how do these patterns differ among sectors?
• How effective are different NLP models in capturing the complex relationships between ESG factors (environmental, social, and governance) and specific financial metrics like profitability, cash flow, and stability?

To address these questions, we offer the following key contributions:

• We develop a novel, multimodel NLP framework that applies nine advanced sentiment analysis models to a large dataset of ESG-related news articles, providing unprecedented depth in understanding the interplay between ESG sentiment and corporate financial performance across multiple industries.
• We uncover and characterize distinct industry-specific patterns in how ESG sentiment affects financial outcomes, demonstrating that the influence of ESG factors is not uniform but varies significantly between sectors, thus emphasizing the importance of customized ESG strategies.
• We critically evaluate the performance of different NLP models in capturing complex ESG–financial relationships, identifying the most effective approaches for nuanced sentiment analysis, and contributing methodological advancements to the field of ESG research.

By integrating a comprehensive cross-industry analysis with advanced NLP techniques, our study offers fresh insights into the multifaceted interactions between ESG sentiment and financial performance, thereby filling critical gaps in the existing literature and informing more effective ESG management strategies [14].

The remainder of this paper is organized as follows: Section 2 provides a review of existing studies on ESG management, sentiment analysis, and its relevance to industries such as mobility, establishing the theoretical framework. Section 3 presents the research findings and discusses key insights regarding ESG management and financial performance across different sectors. Section 4 details the data collection and analysis methodology, introducing the sentiment and financial correlation analysis approaches. Finally, Section 5 summarizes the primary contributions and offers recommendations for future research directions.

2. Related Work

The analysis of ESG factors plays a crucial role in evaluating corporate sustainability and financial performance, and recent studies have increasingly employed NLP techniques to gain deeper insights into ESG data [15]. Existing research primarily focuses on identifying correlations between ESG sentiment and financial performance or examining specific ESG themes using single-model approaches.

Table 1. Overview of prior studies on ESG sentiment analysis and financial correlation in environmental, social, and governance contexts.

Authors	Year	Objective/Purpose	Method and Analytical Tools	Key Findings	Differences from the Present Study
Raman et al. [16]	2020	Analyzed linguistic patterns in ESG topics across corporate earnings calls by industry.	Utilized neural models to classify ESG discourse in earnings calls.	Identified significant industry-specific ESG discourse patterns.	Focused on industry-specific ESG discourse using neural models without a multimodel comparison across ESG factors.
Perazzoli et al. [10]	2022	Examined structural challenges in ESG topics using a systems theory approach.	Applied a systems theory approach to analyze structural ESG issues.	Highlighted challenges within energy and governance themes across industries.	Emphasized single-model analysis of structural ESG issues, lacking multimodel analysis and detailed financial metrics correlation.
Pasch and Ehnes [17]	2022	Enhanced ESG classification performance by fine-tuning transformer models on ESG-related data.	Fine-tuned BERT model on ESG data, achieving higher accuracy.	Achieved 11% higher accuracy than traditional classifiers for ESG sentiment.	Focused on specific model adjustments rather than a comprehensive model comparison across multiple NLP models and industries.
Mehra et al. [11]	2022	Developed an ESG-specific language model to improve document classification accuracy.	Customized an ESG-BERT model fine-tuned on ESG-specific corpora.	Improved accuracy in ESG classification tasks through domain-specific model tuning.	Lacked a multidimensional comparison across ESG domains and did not assess correlations with financial metrics.
Park et al. [18]	2022	Investigated the relationship between public sentiment and corporate resilience using Twitter data.	Analyzed Twitter data to assess public sentiment related to ESG topics.	Found that ESG sentiment on Twitter can be an indicator of corporate resilience during crises.	Utilized a single sentiment analysis model focused on public sentiment without examining detailed financial performance metrics.
Yu et al. [19]	2023	Explored the effect of ESG sentiment on stock price stability.	Examined the correlation between ESG sentiment and stock price volatility.	Identified significant influence of ESG sentiment on stock stability.	Emphasized the correlation with stock stability, limited to specific financial metrics and lacking industry-specific analysis.
Kim et al. [9]	2024	Examined the interconnections between ESG financial trends and sentiment analysis of ESG-related news from 2019 to 2022.	Applied sentiment analysis models to ESG news articles and correlated findings with financial trends.	Identified key relationships between ESG news sentiment and financial performance indicators.	Limited to single-model analysis and specific financial metrics, without a comprehensive cross-industry comparison.
This Study	2024	Utilized nine natural language processing models for ESG sentiment analysis, mapping their relationships to financial performance across industries.	Employed nine NLP models for sentiment analysis of ESG-related news and TF-IDF for key term extraction, examining correlations between sentiment scores and financial performance.	Identified that industry-specific ESG strategies contribute to financial stability, highlighting the importance of ESG practices in sectors like renewable energy and mobility.	Conducted a multimodal comparison, examining diverse correlations between ESG sentiment and detailed financial metrics across multiple industries.

provides a summary of the key contributions from previous studies in ESG sentiment analysis and financial correlation, along with their distinctions from the present study.

Raman et al. [16] used neural models to analyze ESG-related discourse in corporate earnings calls, identifying how ESG issues vary in importance across industries and examining their direct impact on business operations. However, their focus was primarily on industry-level discourse patterns with a limited scope on financial performance implications, leaving the nuanced financial impact of ESG sentiment unexplored. Perazzoli et al. [10], on the other hand, provided a broad literature analysis of ESG issues, covering structural challenges such as energy management, labor practices, and governance ethics. While this study provided a comprehensive view of ESG challenges, it was limited by its qualitative approach, which lacked model-based sentiment analysis and insights into the quantifiable impact on financial metrics.

Pasch et al. [17] and Mehra et al. [11] contributed to the development of ESG-specific NLP models by creating models, such as ESGBERT, tailored to improve classification accuracy and interpretability in ESG contexts. Pasch’s model [17] provided high classification accuracy, while Mehra’s work [11] emphasized interpretability. However, both studies are limited by their single-model focus, as they did not perform comparative analyses across multiple models to assess the variability in ESG sentiment capture. This limits their insights into model performance across ESG themes and sectors.

In contrast, studies by Park et al. [18] and Yu et al. [19] examined the role of ESG sentiment in financial stability, highlighting the link between public ESG sentiment and corporate resilience. Park et al. [18] focused on the role of ESG sentiment in corporate resilience during crises, while Yu et al. [19] emphasized its effect on stock price stability. Despite these valuable findings, both studies relied on limited sentiment analysis models that may not capture the diverse and sector-specific sentiments present in different ESG contexts.

In addition, previous research [9] examined ESG sentiment trends and financial implications from 2019 to 2022 using methodologies such as term frequency-inverse document frequency (TF-IDF), latent dirichlet allocation (LDA) topic modeling, valence-aware dictionary and sentiment reasoner (VADER), and BERT. While this research provided basic insights into how ESG sentiment impacts financial metrics, it was limited by single-model analysis and a limited focus on specific financial metrics, thus lacking a broad, comparative perspective.

In summary, previous research has made significant advances in the field of ESG sentiment analysis, but each study is often limited by reliance on a single model, a narrow thematic or sector focus, or qualitative approaches that lack quantifiable insights into financial performance. Most studies emphasize either the development of specific NLP models or the examination of isolated ESG issues, resulting in a partial understanding of ESG impacts.

In contrast, our study adopts a comprehensive multimodel approach, applying nine advanced NLP models, including BERT and RoBERTa, to conduct a thorough analysis of ESG sentiment across multiple industries. This comparative model approach not only allows us to evaluate the effectiveness of different models but also enables a nuanced exploration of the relationships between ESG sentiment and various financial performance indicators across industries.

By including different industries and multiple sentiment models, our study captures the multifaceted impact of ESG on corporate sustainability and fills gaps in model comparison, industry-specific analysis, and the financial impact of ESG sentiment. This research thus fills a critical gap by providing a cross-sector, model-comparative analysis that enhances our understanding of how ESG sentiment impacts financial performance in different contexts. Through this innovative approach, we contribute practical guidance for industry-specific ESG strategy development and a more refined theoretical framework that incorporates sentiment as a central component in assessing ESG impacts on financial sustainability.

3. Methodology

This study takes a quantitative research approach, using statistical and machine learning techniques to analyze the correlation between ESG sentiment and financial performance across industries. By adopting this framework, we aim to provide objective and measurable insights into the impact of ESG sentiment on corporate financial outcomes. Specifically, we apply several NLP models, including advanced models such as BERT, RoBERTa, and so on, to perform sentiment analysis on ESG-related news and examine how sentiment scores correlate with financial indicators across industries. Our methodology consists of three main steps: identifying key ESG themes with TF-IDF analysis, applying multimodel NLP approaches to sentiment classification, and performing correlation analysis to quantify relationships between sentiment scores and financial metrics such as profitability, cash flow, and stability. This structured, data-driven approach enables a robust exploration of ESG impacts across sectors, providing comprehensive cross-industry insights into ESG-related financial dynamics.

3.1. Identifying Key ESG Topics with TF-IDF Analysis

To assess the importance of specific ESG-related keywords in headlines and leadlines, we used the TF-IDF method, which calculates the relevance of each word based on its frequency in the text and rarity across the dataset to accurately extract primary topics [20].

• We cleaned the text data by removing unnecessary symbols, numbers, and stopwords. Stemming and lemmatization were applied to maintain word consistency. This preprocessing improves the quality of the data and ensures that the analysis focuses on meaningful and relevant terms.
• TF-IDF weights were calculated for the preprocessed data, reflecting each word’s relative importance in the document. This helped identify the frequency and relevance of the ESG keywords, transforming them into high-weight terms representative of the primary topics.
• Words with high TF-IDF weights were extracted to identify ESG-related themes and corporate reputational topics and served as the basis for subsequent analyses. This extraction allows for a focused examination of the most important ESG issues affecting company performance.

3.2. Sentiment Classification Across Multimodel NLP Approaches

We employed a diverse set of sentiment analysis models to classify the emotional direction of ESG news, employing nine models (BERT, RoBERTa, BigBird, DistilBERT, ALBERT, TinyBERT, ELECTRA, VADER, and TextBlob) to categorize each news item’s sentiment as negative, neutral, or positive. To improve efficiency, lightweight models such as DistilBERT, ALBERT, and TinyBERT were applied with binary (positive/negative) classification capabilities, and ELECTRA was included as an additional binary model due to its high performance and fast processing speed in distinguishing fine-grained differences between sentiments.

Larger models such as BERT, RoBERTa, and BigBird, trained on extensive datasets, provide robust contextual understanding, which improves sentiment classification accuracy across multiclass categories [21]. Conversely, the lightweight models, i.e., DistilBERT, ALBERT, and TinyBERT, allow rapid binary classification across large datasets without sacrificing significant accuracy. For consistency, each model was loaded with pretrained parameters, and the prediction classes were determined based on the model output logits. This approach allows for the evaluation of trade-offs between NLP models, supporting informed model selection for future ESG sentiment analysis.

3.2.1. Multiclass Sentiment Analysis Models

• BERT [22]: BERT is a transformer-based model that analyzes text in both directions to understand the context. The model can identify the meaning of words in a sentence through interactions between words. In this study, we chose Bert-base-cased among various BERT models [23].
• RoBERTa [24]: RoBERTa is a model that performs optimized learning on larger datasets based on the underlying structure of BERT. This model is designed to provide a more precise understanding of various linguistic characteristics. In this work, we used the cardiff-nlp/twitter-roberta-base-sentiment model, which boasts high accuracy, especially in emotion analysis [25].
• Big BERT (BigBird) [26]: BigBird was developed to overcome the limitations of conventional transformer models and effectively handle long texts. Using Google/bigbird-roberta-base, this model can identify emotions even when analyzing long and complex ESG news stories without compromising the context. It is particularly effective in emotion analysis in long texts.
• VADER [27]: VADER is a rule-based emotional model suitable for analyzing informal or unstructured text. VADER is particularly strong on informal text, such as social media, and can quickly derive emotional results. This capability enables fast, effective, real-time sentiment analysis of user content.
• TextBlob [28]: TextBlob is a rule-based emotion classification tool that quickly classifies emotions into positives, neutralities, and negatives. In this study, this tool served as a baseline for emotion analysis and provided basic data for comparing performance with pretrained models with VADER.

3.2.2. Binary-Class Sentiment Analysis Models

• Distilled BERT (DistilBERT) [29]: DistilBERT is a lightweight BERT model that provides a faster inference speed while maintaining the performance of BERT. In this study, we chose this model for efficient emotion analysis on large news datasets.
• A Lite BERT (ALBERT) [30]: ALBERT is a model designed to achieve faster processing speeds by reconstructing the parameter structure of BERT. The model is suitable for binary classification in emotion classification tasks, allowing for the fast classification of affirmations or negatives.
• Tiny BERT (TinyBERT) [31]: TinyBERT is a model designed to further reduce the architecture of BERT. The model enables efficient binary emotion classification and performs well in distinguishing between positive and negative emotions.
• Efficiently Learning an Encoder that Classifies Token Replacements Accurately (ELECTRA) [32]: ELECTRA reduces computational demands using an alternative token prediction method. This model was employed to enhance binary classification performance, specifically in differentiating positive and negative sentiments.

The performance of each model was evaluated in terms of sentiment classification accuracy, speed, and efficiency, allowing us to identify the best models for ESG-related predictions. In addition, we visualized model prediction patterns using uniform manifold approximation and projection (UMAP), a dimensionality reduction technique that simplifies high-dimensional data for better visualization, to understand high-dimensional embedding data across different models [33,34]. This approach enabled us to examine the patterns and variations in sentiment predictions across models, enhancing the interpretability and consistency of the sentiment classification results.

3.3. Correlation Analysis

To explore the relationships between ESG sentiment scores and financial performance across industries, we conducted a correlation analysis to quantify the relationships between sentiment categories (i.e., positive, neutral, and negative) and financial metrics, including profitability, cash flow, and stability. This structured approach facilitates an objective examination of how ESG sentiment can influence financial outcomes across sectors.

• Data aggregation and sector classification: sentiment scores were first categorized by sentiment type and then aggregated by industry to enable sector-specific analysis [35]. In this step, we calculated average sentiment scores for each article within a sector and compared these scores to the financial indicators of companies in that sector. Aggregating the data at the industry level allowed us to capture unique sectoral patterns and interpret the varying impact of ESG sentiment across industries, such as mobility and renewable energy.
• Analytical approach: in this correlation analysis, we compared aggregated sentiment scores to corresponding financial indicators, using a consistent correlation coefficient to measure the alignment between sentiment and financial metrics [36]. This approach provided a robust framework for examining how changes in sentiment may be correlated with financial changes within each sector. By identifying these relationships, we aimed to provide a systematic view of how ESG sentiment correlates with key financial metrics across industries.

This correlation analysis framework thus provides the basis for interpreting the sector-specific findings in the Results section, allowing us to explore the relevance of ESG sentiment to financial outcomes on an industry-specific basis.

4. Data Acquisition and Preparation

This section is divided into two parts. Section 4.1 outlines the methodologies and sources used to obtain ESG-related news and financial data. Section 4.2 describes the procedures employed to purify and consolidate the data for analytical purposes.

4.1. Data Collection

To examine global trends in ESG management, we used the Google News platform with English as the language setting and “ESG finance” as the primary search term [37]. We chose “ESG finance” to capture articles that specifically discuss the intersection of ESG issues and corporate financial performance, ensuring that the collected articles were highly relevant to our research objectives. In the initial data collection phase, all articles matching this keyword were collected without additional filtering or categorization by type (e.g., hard news, editorials, and opinion pieces), which allowed us to collect relevant online news articles from a broad, international perspective for contextual analysis of ESG sentiment related to corporate financial performance [38]. The collection period spanned from June 2019 to May 2022 and was chosen to capture changes in ESG activity and sentiment before, during, and after the COVID-19 pandemic, allowing for analysis of changes in ESG-related news coverage over time [39]. We used a web crawling approach, dividing the collection process by month to minimize data loss associated with large-scale web crawling [40].

This method yielded a total of 9828 English-language news articles, including a variety of article types such as opinion pieces, editorials, and hard news, all written in English, which eliminated the need for translation and reduced the risk of meaning loss that can arise during translation processes. This mix of article types, each with its own unique structure and tone, allowed for more nuanced sentiment analysis that fully represented different perspectives. Recognizing the differences in structure and tone among these types, we acknowledged these variations during our analysis to ensure a comprehensive representation of sentiment across all article categories. For sentiment analysis, we focused primarily on the “headline” and “lead” sections of each article, as these sections typically summarize the main content and sentiment of the article. This focus on headlines and leads allowed us to capture key points that reflect public sentiment on ESG issues, while minimizing the inclusion of general background information. Although we did not explicitly follow the journalistic “inverted pyramid” structure, we prioritized these sections because they are generally the most impactful for understanding sentiment, which is particularly relevant for readers familiar with journalistic practices. This targeted approach allowed us to maintain relevance and clarity by prioritizing the sections most relevant to understanding sentiment.

Sentiment classification was performed using a variety of NLP models to maintain consistency and objectivity across the dataset. For multiclass classification, BERT, RoBERTa, and BigBird were used to classify sentiment into positive, neutral, and negative classes. For binary classification, DistilBERT, ALBERT, TinyBERT, and ELECTRA were used to efficiently distinguish positive from negative sentiment. In addition, VADER and TextBlob provided rule-based sentiment scoring for fast and consistent analysis, especially for unstructured text. By employing these predefined model functions and standardized criteria, sentiment interpretation remained consistent; thus, an additional interrater reliability test was deemed unnecessary due to the automated nature of these models. In addition, this foundational dataset was complemented with news articles specifically related to individual companies’ ESG activities to investigate the impact of ESG efforts on financial performance. Throughout the data collection process, we monitored for special cases such as duplicate articles or irrelevant content, addressing them by removing duplicates and filtering out non-relevant articles to enhance the dataset’s quality and representativeness. Although the dataset may not encompass every article on corporate financial performance, it provides a substantial and representative cross-sectional view of ESG narratives relevant to financial outcomes over the specified period.

For the company-specific news analysis, we selected 773 U.S.-listed companies with ESG ratings from MSCI [41]. These companies are listed on major exchanges, including the (NYSE), NASDAQ, and AMEX. News articles for each company were gathered from Google News using a combination of the company name and “ESG” as search terms (e.g., “Apple ESG”). This approach was designed to capture a comprehensive view of the ESG-related media coverage. To ensure thorough data acquisition, we employed Selenium WebDriver and ChromeDriver for dynamic web content crawling, retrieving up to 30 pages of news articles per company [42]. Through this process, we acquired over 140,000 articles, which formed a rich dataset on company-specific ESG activities. These data points provided a wide-ranging view of each company’s ESG-related actions, which we subsequently analyzed to discern the sentiment direction and examine its correlation with financial performance.

To analyze the financial performance of each company, financial data spanning from 2019 to 2021 were collected, corresponding to the news data collection period to maintain temporal consistency and improve analytical accuracy. Financial data were obtained from Yahoo Finance, focusing on key indicators that provided a comprehensive assessment of a company’s profitability, cash flow, and stability [43]. Specifically, for profitability, we gathered revenue, revenue growth rate, and return on assets (ROA). For cash flow, we included earnings before interest, taxes, depreciation, and amortization (EBITDA) and its growth rate. Finally, for stability, we incorporated interest expense, interest expense growth rate, and debt-to-equity ratio. These indicators collectively provided insights into each company’s financial health, enabling a multidimensional view of financial performance.

Data extraction from Yahoo Finance was accomplished using the BeautifulSoup library (version 4.12.2) to parse HTML content and extract relevant information systematically [44]. Each company’s financial data were then organized for analysis, aligning financial and sentiment data temporally by year to ensure consistency. To assess ESG management performance, we collected ESG rating data from MSCI, including an overall ESG score and specific scores for ESG aspects [45]. These rating data served as an essential benchmark, providing a professional assessment of a company’s ESG activities, which we compared against public sentiment derived from news data. Analyzing these ratings relative to sentiment data allowed us to investigate the alignment between public perception and professional ESG evaluations, potentially uncovering discrepancies that may impact investment decisions.

4.2. Data Preprocessing and Integration

Once the data were collected, we performed comprehensive text preprocessing to enhance the quality and relevance of the dataset for sentiment analysis. First, we standardized all text data by converting them to lowercase letters, which reduced case-based inconsistencies [46]. We then employed regular expressions to remove punctuation, numbers, special characters, and non-English characters, thereby ensuring a focus on meaningful words [47]. To address abbreviations, common contractions (e.g., “don’t”) were expanded to their full forms (e.g., “do not”) to preserve semantic accuracy. Stopwords, which carry minimal semantic value, were removed using the Natural Language Toolkit (NLTK, version 3.9.1) stopword dictionary [48]. Tokenization was performed via NLTK’s word_tokenize function, segmenting the text into individual words for more granular analysis [49]. Finally, where appropriate, stemming or lemmatization was applied to consolidate word forms, grouping variations (e.g., “run” and “running”) under a single root word [50]. This preprocessing stage was vital in enhancing the analytical robustness of the dataset, allowing for more reliable and accurate insights into ESG sentiment trends.

For comparative analysis, the financial and sentiment datasets were integrated at the company level and organized by year. For example, news data from 2020 were analyzed, along with financial data from the same year, facilitating a direct examination of contemporaneous financial and ESG sentiment correlations. Financial metrics were normalized using relative measures, such as revenue growth and EBITDA growth, to account for company size differences and improve cross-company comparability. Recognizing the inherent differences in financial structure across industries, we classified companies into 11 sectors according to the Sustainable Industry Classification System developed by the Sustainability Accounting Standards Board [51,52]. This sectoral categorization enabled industry-specific analysis, accounting for variations in ESG priorities and financial structures that may influence ESG–financial performance relationships. Through these meticulous data acquisition and preparation steps, we established a structured and reliable dataset that formed the foundation for examining the interactions between ESG sentiment, professional ESG assessments, and financial performance.

5. Results and Discussion

This section presents an analysis of the results of our comprehensive study of ESG-related news text. Using TF-IDF analysis, we identified key ESG-related keywords over time, revealed shifts in focus within ESG management, and explored the intricate links between ESG issues and the financial and operational dimensions of companies. For example, the predominance of neutral sentiment observed in the multiclass sentiment analysis is consistent with the findings of Yu et al. [19], who highlighted a trend toward neutrality in ESG news coverage. Our use of both multiclass and binary models allows us to capture more detailed sentiment dynamics across industries, extending the findings of Yu et al. by providing a cross-industry perspective on how neutral sentiment correlates with financial outcomes. In addition, the relationship between neutral sentiment and positive financial outcomes echoes Perazzoli et al.’s [10] findings on how public sentiment can influence corporate reputation, suggesting that balanced or neutral ESG sentiment may promote financial stability by fostering trust and reliability. By linking these findings to previous studies, we provide a broader context for understanding the evolving role of ESG sentiment in financial and operational performance.

5.1. TF-IDF Analysis

First, we performed TF-IDF analysis of the top keywords extracted from headlines and lead text in ESG-related news stories over the entire research period from June 2019 to May 2022.

Table 2. Summary of key ESG keywords and their TF-IDF weights (June 2019–May 2022).

Rank	Headlines		Leads
	Keyword	TF-IDF	Keyword	TF-IDF
1	Autos	1	disillusionment	0.886
2	Doctor	1	risk	0.872
3	esg	1	bonds	0.854
4	Green	1	award	0.854
5	greenium	1	excellence	0.820
6	Horizon	1	jbs	0.816
7	Illustrated	1	bgc	0.816
8	Mercy	1	investing	0.812
9	Nordics	1	servicenow	0.808
10	Revealed	1	Loans	0.804
11	Talk	0.990	tigo	0.799
12	antitrust	0.978	materiality	0.798
13	epicenter	0.978	Xylem	0.785
14	abc	0.976	abaxx	0.783
15	operationalize	0.976	ci	0.782

shows that environmental terms such as “green”, “greenium”, and “horizon” ranked prominently, highlighting the centrality of environmental concerns in ESG management. Notably, financial terms such as “loans”, “bonds”, “antitrust”, and “investing” also highlight the financial interlinkage in ESG efforts, indicating that ESG management is deeply intertwined with corporate financial practices. This analysis demonstrates that effective ESG strategies integrate environmental and financial practices, which are indispensable for the optimal functioning of a corporation.

5.1.1. Year-by-Year Analysis of Headlines

The year-by-year breakdown of headline analysis in

Table 3. Annual trends in ESG keyword emphasis based on headline analysis.

Rank	June 2019–May 2020.05		June 2020–May 2021		June 2021–May 2022
	Keyword	TF-IDF	Keyword	TF-IDF	Keyword	TF-IDF
1	autos	1	esg	1	esg	1
2	doctor	1	greenium	1	horizon	1
3	green	1	mercy	1	abc	0.981
4	illustrated	1	nordics	1	antitrust	0.981
5	revealed	1	talk	0.987	putting	0.977
6	epicenter	0.965	crossroads	0.963	tracker	0.977
7	fails	0.962	dna	0.963	accountants	0.974
8	rip	0.959	operationalize	0.963	ready	0.971
9	primer	0.957	initiative	0.954	way	0.967
10	decade	0.955	importance	0.947	war	0.946

illustrates how the focus of ESG management evolved annually. From June 2019 to May 2020, terms such as “epicenter”, “fails”, and “rip” were prevalent, along with “green”, reflecting initial uncertainty and negative perceptions about early ESG management. Between June 2020 and May 2021, keywords such as “greenium”, “operationalize”, “initiative”, and “importance” became prominent, marking a phase in which ESG management became mainstream. Finally, between June 2021 and May 2022, terms such as “horizon”, “antitrust”, and “tracker” dominated, suggesting that ESG practices were now viewed as not merely initiatives but as impactful factors influencing corporate financial outcomes.

5.1.2. Year-by-Year Analysis of Leads

As shown in

Table 4. Annual trends in ESG keyword emphasis based on lead analysis.

Rank	June 2019–May 2020		June 2020–May 2021		June 2021–May 2022
	Keyword	TF-IDF	Keyword	TF-IDF	Keyword	TF-IDF
1	disillusionment	0.874	risk	0.878	award	0.857
2	materiality	0.823	bonds	0.873	servicenow	0.793
3	investing	0.805	loans	0.820	jbs	0.793
4	tigo	0.782	excellence	0.816	director	0.781
5	director	0.775	director	0.783	xylem	0.781
6	trade	0.737	citi	0.769	bgcr	0.761
7	msci	0.734	ci	0.762	lgbtq	0.754
8	ocean	0.731	keamy	0.759	abaxx	0.750
9	stocks	0.718	vale	0.756	wanda	0.749
10	spy	0.717	regulations	0.741	assurance	0.749

, the year-by-year analysis of the lead text revealed governance and financial themes within ESG as major topics, similar to the headlines. From June 2019 to May 2020, governance-related terms such as “disillusionment”, “materiality”, and “investing” featured prominently, indicating the prevalence of governance discussions in the early stages of ESG. Between June 2020 and May 2021, financial terms such as “bonds” and “loans” were dominant, suggesting an increasing intersection between finance and ESG aligned with the emergence of green bonds and ESG investing. This shift underscores the integration of ESG into finance and points to a trend toward sustainable investing.

This period also saw terms such as “regulations” reflecting heightened discussions on environmental regulation and sustainability legislation. Between June 2021 and June 2022, with the occurrence of the inaugural ESG awards, corporate names (such as ServiceNow, JBS, Xylem, and Abaxx) emerged more frequently than specific keywords related to environment, society, and governance. Social terms such as “LGBTQ” also first appeared, highlighting the expanded scope of ESG into social responsibility. This expansion demonstrates that modern ESG practices are becoming more comprehensive, addressing a broad range of social and corporate responsibilities.

5.1.3. Industry Sentiment Trends in ESG Keywords

The results are consistent with previous studies on the importance of environmental and financial terms in ESG-related discourse. For example, Raman et al. [16] analyzed corporate earnings calls and found a similar emphasis on environmental and financial terms, which is consistent with our findings, where keywords such as “green”, “greenium”, and “bonds” consistently ranked high in TF-IDF scores. This suggests a shared focus in corporate and public discourse on the financial and environmental aspects of ESG, particularly in industries heavily influenced by sustainability demands.

In addition, Pasch and Ehnes [17] highlighted the benefits of fine-tuning NLP models to improve performance in ESG-related text classification. We adopted this approach to improve accuracy in capturing critical ESG terms. While Pasch’s study demonstrated model effectiveness in a specific ESG context, our broader application across multiple industries underscores an even more robust ability to capture ESG themes through TF-IDF analysis. This cross-sector perspective goes beyond the work of Pasch and Ehnes [17] to provide a more comprehensive view of how ESG keywords differ across industries and their impact on financial performance.

In contrast to previous single-focus studies, this analysis incorporates multiple ESG dimensions to account for different sector impacts. For example, the increased prominence of terms such as “operationalize” and “initiative” between 2020 and 2021 echoes the findings of Yu et al. [19], who found that ESG management initiatives are gradually gaining mainstream acceptance. However, our study goes beyond Yu et al. by examining how specific terms evolve and correlate with financial indicators over multiple years, revealing industry trends in response to ESG pressures.

To further interpret these findings, it is important to highlight shifts in sentiment and keyword prominence over time. For example, the rise of terms such as “antitrust” and “tracker” in the later period (2021–2022) reflects a shift in the industry to view ESG not only as a set of initiatives but also as an essential driver of corporate accountability and performance. This shift is consistent with Perazzoli et al. [10], who argued that public sentiment has a significant impact on corporate reputation. Our year-over-year keyword analysis supports this perspective, showing an increasing public and financial focus on accountability through keywords such as “antitrust” and “tracker”.

5.2. Sentiment Analysis

In this study, we leveraged various sentiment analysis models to assess sentiment in ESG-related news headlines and leads. Multiclass models (BERT, RoBERTa, VADER, TextBlob, and BigBird) were used to classify sentiment as negative, neutral, or positive, and binary classification models (DistilBERT, ALBERT, TinyBERT, and ELECTRA) categorized sentiment as either positive or negative.

5.2.1. Multiclass Sentiment Analysis of Headlines

For the headline analysis, we employed multiclass sentiment models, namely, BERT, RoBERTa, VADER, TextBlob, and BigBird, to classify sentiment into negative, neutral, and positive categories. As shown in Figure 1, neutral sentiment was the most common across the models, followed by positive sentiment, with negative sentiment being the least common. Notably, the RoBERTa, TextBlob, and BigBird models exhibited a stronger tendency toward neutral sentiment, indicating a generally neutral tone in ESG news coverage, with an occasional positive emphasis.

5.2.2. Binary Sentiment Analysis of Headlines

Binary classification models, namely, DistilBERT, ALBERT, TinyBERT, and ELECTRA, were applied to classify ESG news headlines as positive or negative. As shown in Figure 2, DistilBERT, ALBERT, and TinyBERT exhibited a relatively balanced distribution of positives and negatives, suggesting that they reflect the overall tendency of the text in the context of being classified solely as positive or negative. Conversely, ELECTRA detects negative emotions at a high rate, indicating that different models yield varying emotional outcomes depending on the overall tone of the data.

5.2.3. Multiclass Sentiment Analysis of Leads

Applying multiclass models to leads revealed a sentiment distribution similar to that of the headlines. As shown in Figure 3, VADER and TextBlob exhibited stronger inclinations toward positive sentiment, whereas BERT and RoBERTa retained higher proportions of neutral sentiment. However, BigBird recorded the highest levels of positive sentiment in the lead text, likely reflecting the additional context within the leads that emphasizes positive ESG-related content.

5.2.4. Binary Sentiment Analysis of Leads

The binary sentiment analysis of leads using DistilBERT, ALBERT, TinyBERT, and ELECTRA mirrored the trends observed in headlines. As shown in Figure 4, the ELECTRA model once again exhibited a higher tendency to classify content as negative, possibly because of a lower threshold for neutral categorization. In contrast, DistilBERT and ALBERT demonstrated a relatively balanced sentiment distribution, suggesting that the additional context in the leads typically highlights more positive information.

5.3. Visualize the Financial Link to ESG

This section presents a UMAP visualization to intuitively illustrate the relationships between ESG-related keywords and their financial and environmental implications. UMAP, known for its efficiency in handling large datasets, excels at simplifying complex, high-dimensional data into manageable, low-dimensional representations. This allows us to capture the non-linear data structures in ESG news content and highlight clustering patterns among the extracted keywords.

In this analysis, the UMAP parameters were optimized to enhance the clarity of keyword clustering related to ESG news. Specifically, the n_neighbors parameter was set to three to emphasize local connectivity among ESG and financial keywords, creating tightly clustered groups within specific topics by focusing on the top 100 keywords [53]. In addition, a low min_dist value of 0.01 was selected to encourage close linkages among keywords, revealing distinct clusters that represent various ESG themes.

• The headline UMAP analysis (Figure 5) reveals a clear distribution of keywords across environmental and financial themes. Words such as “sustainable”, and “green” are positioned closely to financial terms such as “earnings”, “shares”, and “stocks”, suggesting a close association between ESG management and financial performance. This clustering visually highlights how ESG-related environmental factors are directly connected to corporate financial operations, providing an intuitive overview of their interrelations.
• In the lead UMAP analysis presented in Figure 6, keywords related to ESG and socially responsible management are prominently clustered, along with financial terms such as “equity”, “funds”, and “assets”. This arrangement suggests a potential link between the social and governance aspects of ESG and long-term financial outcomes. For example, keywords such as “sustainability”, “social”, and “governance” align closely with financial terms, indicating that firms prioritizing social responsibility experience positive financial performance.

5.4. Correlation Analysis of ESG Sentiment and Financial Performance

This study analyzes the relationship between sentiment analysis data, ESG ratings (provided by MSCI), and industry-specific financial metrics, focusing on profitability, cash flow, and stability. The primary aim is to identify how ESG-related news sentiment correlates with corporate financial performance across different industries. The analysis explores the impact of ESG ratings on financial indicators and investigates how sentiment analysis results relate to the financial outcomes of companies.

Financial factors are categorized into three areas:

• Profitability: assessed using revenue (2021), revenue growth rate, and ROA (2021) to evaluate a company’s earning power. The year 2021 was chosen as the reference year to observe changes in profitability following the pandemic period, allowing a focused analysis of ESG sentiment in the post-pandemic economic environment.
• Cash flow: evaluated through EBITDA and its growth rate (2021), reflecting net cash flow from operations and its growth. Using 2021 data provides insight into operational cash flow stability and growth during the recovery phase post-pandemic.
• Stability: measured using interest expense, interest expense growth rate, and debt-to-equity ratio (2021) as key indicators of financial strength. These metrics indicate a company’s ability to manage debt and operational leverage in response to the changing economic conditions after the pandemic.

The analysis spans three years, from 2019 to 2021, with the correlation results visualized through heatmaps in Appendix A (Figure A1, Figure A2, Figure A3, Figure A4, Figure A5, Figure A6, Figure A7, Figure A8, Figure A9, Figure A10 and Figure A11). The choice of 2021 as the reference year for these financial metrics is intentional, as it allows us to assess the impact of ESG sentiment on financial performance during a significant post-pandemic economic transition. For Figure A1, Figure A2, Figure A3, Figure A4, Figure A5, Figure A6, Figure A7, Figure A8, Figure A9, Figure A10 and Figure A11, each financial variable is coded based on 2021 values to maintain consistency across industries and allow for a clear identification of post-pandemic trends. This approach highlights recent changes in financial performance as companies adapt their ESG practices in response to evolving stakeholder expectations in the post-pandemic period. Growth rates, such as revenue growth and EBITDA growth, reflect annual changes from 2020 to 2021, allowing for year-over-year comparisons in the context of recovery. The ROA metric specifically refers to 2021 values as it provides a focused measure of asset efficiency following the economic disruptions of the pandemic.

The results show different correlations between ESG factors and financial indicators across industries, reflecting the unique characteristics of each sector.

Table 5. Summary of ESG sentiment and financial correlations by industry sector.

Industry	Key Findings	Figure Reference
Consumer Goods	Positive sentiment; low environmental score; favorable financial indicators	Figure A1
Extractives and Minerals Processing	Strong ESG–governance correlation; profitability; cash flow	Figure A2
Financial	Mixed sentiment effects; positive ESG–financial performance correlation	Figure A3
Food and Beverage	Positive sentiment with ESG; low financial correlation	Figure A4
Healthcare	Positive governance sentiment; limited financial impact	Figure A5
Infrastructure	High environmental score; positive financial correlation; leverage influence	Figure A6
Renewable Resources and Alt. Energy	Strong environmental–financial stability correlation	Figure A7
Resource Transformation	Minimal ESG–financial correlation; high governance score	Figure A8
Services	Varied impacts by service type; mixed ESG correlations	Figure A9
Technology and Communications	Positive environmental–profitability correlation; limited governance impact	Figure A10
Transportation	High ESG–sentiment correlation; environmental sensitivity	Figure A11

below provides a consolidated summary of the key findings from our correlation analysis, allowing readers to quickly grasp the key insights. Each industry shows varying degrees of sensitivity to ESG factors; for example, sectors such as mobility and renewable energy show significant impacts from environmental sentiment, while others such as healthcare and financial services show more nuanced relationships. This table allows readers to efficiently locate key findings across sectors. A detailed analysis and visualizations for each sector can be found in Appendix A. These figures (Figure A1, Figure A2, Figure A3, Figure A4, Figure A5, Figure A6, Figure A7, Figure A8, Figure A9, Figure A10 and Figure A11) provide in-depth visual insights, including heatmaps and correlation matrices, to illustrate the unique relationships between ESG sentiment and financial performance within each industry.

5.5. Discussion on Industry-Specific ESG Sentiment and Financial Impacts

This study conducted a sentiment analysis of ESG-related news headlines and leads, examining the relationship between sentiment, company financial performance, and ESG ratings across various industries, including the mobility sector. The analysis highlights correlations between ESG news sentiment and company profitability, cash flow, and stability metrics. The findings indicate that neutral or positive ESG news sentiment is often associated with positive financial outcomes, particularly in sectors where environmental and social factors significantly influence company perceptions, such as mobility and renewable energy. This trend underscores the perspective that positive public sentiment toward ESG can promote corporate resilience and alignment with societal expectations for sustainable practices, thereby fostering accountability and trust among stakeholders. These findings are consistent with previous studies showing that macroeconomic factors and news sentiment significantly influence stock returns [54]. Similarly, social media sentiment around major political events has been found to influence stock markets [55]. The results suggest that the relationship between sentiment and financial performance varies by industry, with environmental and social factors having a significant impact in sectors such as mobility and renewable energy while having a limited impact in other sectors.

From a literature perspective, this study contributes to the growing body of knowledge on ESG sentiment analysis by employing both multiclass and binary models to provide nuanced sentiment metrics across industries. This dual-model approach enhances the depth of ESG analysis and provides new insights into sector-specific sentiment trends, complementing the existing research that typically uses single-model approaches. The link between positive sentiment and positive financial results supports the view that alignment with societal ESG expectations can strengthen both corporate reputation and financial stability. These industry-specific findings reveal distinct patterns of ESG sentiment across industries and complement the work of Park et al. [18], who found that positive public ESG sentiment is correlated with corporate resilience. This study builds on this foundation by examining additional financial metrics, such as cash flow and stability, to provide a deeper understanding of how ESG sentiment influences financial outcomes. In addition, we extend the work of Mehra et al. [11] on the ESGBERT model for improved ESG context capture through our multimodel approach, which improves the accuracy and interpretability of sentiment analysis across different ESG issues. This multimodel comparison addresses the single-model limitation identified in previous research and provides a more comprehensive view of the financial impact of ESG. For example, in the consumer discretionary sector, positive sentiment correlates with strong financial performance despite lower environmental ESG scores, while in the mobility sector, favorable environmental factors directly contribute to financial gains.

From a practical perspective, the use of both multiclass and binary sentiment models allowed for nuanced sentiment metrics, providing deeper insights into the complex impact of ESG sentiment on financial performance that can guide companies in tailoring their ESG practices. The multiclass model identified a high proportion of neutral sentiment, suggesting that ESG news is generally positive or neutral, while the binary model highlighted polarized financial impacts by distinguishing between positive and negative sentiment. This alignment of neutral or positive sentiment with financial stability reflects broader societal priorities for ethical corporate behavior and long-term stability, which is particularly valuable for practitioners seeking to maintain public trust and sustainable growth [56]. The approach is consistent with studies using advanced machine learning techniques for automated stock market trading, demonstrating the value of sophisticated models in financial decision-making. In addition, the TF-IDF analysis uncovered frequently mentioned ESG-related keywords, illustrating the varying importance of ESG factors across industries. Similar to research using machine learning in demand forecasting, this industry-specific analysis improves the effectiveness of ESG strategies by focusing on relevant ESG factors in each sector, helping practitioners develop targeted, impactful strategies [57].

From a societal perspective, these findings underscore the importance of transparent ESG reporting and management for positive public perceptions that can enhance corporate reputation and align with societal goals for sustainability. Such alignment with public expectations for accountability and ethical standards is critical as it builds trust and supports the pursuit of long-term sustainable practices [58]. This highlights the urgent need for companies to tailor their ESG strategies to the specific needs of their industries, particularly in high-impact sectors such as mobility. In addition, the use of advanced machine learning techniques, such as graph neural networks, offers a promising way to improve the predictive power of ESG analysis and provide more accurate and industry-relevant insights [59]. Taken together, these considerations underscore the urgent need for a forward-looking, nuanced approach to ESG management that addresses both societal expectations and industry-specific challenges.

6. Conclusions

This study provides comprehensive insights into the relationship between ESG news sentiment and corporate financial performance, with a particular focus on industries such as mobility. Using both multiclass and binary classification models, we examined how ESG news sentiment affects key financial metrics, including profitability, cash flow, and stability, highlighting the potential for effective ESG management to improve corporate outcomes. Our findings highlight industry-specific variations in the correlation between ESG sentiment and financial performance, underscoring the need for ESG strategies tailored to the unique characteristics of each sector, particularly dynamic sectors such as mobility.

The key findings of this study are as follows:

• Industry-specific effects: different industries show different levels of correlation between ESG sentiment and financial performance. Sectors such as mobility and renewable energy are particularly affected by environmental sentiment, indicating their heightened sensitivity to ESG news and its impact on company results.
• Modeling approach: the use of both multiclass and binary sentiment models allowed for a nuanced analysis of ESG sentiment. The models revealed a high proportion of neutral sentiment in general ESG news while also highlighting the distinct impact of polarized sentiment on financial performance.
• Strategic implications: developing ESG strategies tailored to the unique characteristics of each industry can improve long-term company performance. This is particularly relevant for sectors that are more sensitive to ESG factors, where tailored approaches can better support sustainable growth and stakeholder trust.

For practitioners, this study provides valuable strategic insights for companies seeking to improve long-term performance through tailored ESG management. A key contribution of this study is to demonstrate the feasibility of using sentiment analysis to assess the impact of ESG initiatives. By using sentiment analysis models to examine ESG sentiment indicators across industries and their correlations with financial performance, our findings show that integrating ESG ratings with sentiment analysis can serve as a reliable predictor of corporate outcomes.

In terms of theoretical implications, this study makes several contributions to the ESG and finance literature. First, it demonstrates that ESG sentiment analysis can reveal industry-specific nuances in the relationship between ESG activities and financial outcomes, providing a customized perspective that differs from traditional ESG scoring approaches that may overlook industry-specific factors. This finding extends existing theories of stakeholder perception and corporate performance by incorporating sentiment as a key component in assessing ESG impacts, thereby broadening the theoretical understanding of how sentiment affects corporate success in different contexts.

However, it is important to acknowledge certain limitations of this study. First, the data used for the sentiment analysis were limited to ESG-related news headlines and leads, which may not capture all the nuances of companies’ ESG practices. In addition, the observed correlation between ESG sentiment and financial performance does not imply causation, as this analysis relied on contemporaneous data rather than longitudinal datasets. These limitations underscore the need for cautious interpretation of the results and highlight areas for improvement in future research.

To address these limitations, future research should focus on expanding data collection efforts to include a broader range of sentiment sources, such as social media and analyst reports, to provide a more comprehensive view of ESG sentiment. In addition, collecting longitudinal data over longer periods of time would allow researchers to more effectively establish causal relationships between ESG sentiment and financial performance. By incorporating these advanced methodologies, future studies can develop more robust and comprehensive models for ESG analysis, increasing the reliability of insights and contributing to better-informed investment decisions and corporate strategies.

In conclusion, this study presents an analytical methodology that is of practical value to companies developing ESG strategies and represents a significant advance in understanding the complex interplay between ESG sentiment and financial performance, particularly in the mobility industry. Through this research, we highlight the role of sentiment analysis in improving ESG assessments across sectors, ultimately benefiting academia, society, and industry practitioners alike.