Next Article in Journal
A Novel Fully Digital Feedforward Background Calibration Technique for Timing Mismatch in M-Channel Time-Interleaved ADCs
Previous Article in Journal
Lightweight Deep Learning Model for Real-Time Colorectal Polyp Segmentation
Previous Article in Special Issue
Deep Feature Extraction for Detection of COVID-19 Using Deep Learning
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Electronics
Volume 12
Issue 9
10.3390/electronics12091964
electronics-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Content and Sentiment Analysis of The New York Times Coronavirus (2019-nCOV) Articles with Natural Language Processing (NLP) and Leximancer

by
Sezai Tunca
*,
Bulent Sezen
and
Yavuz Selim Balcioglu
Management Information System Department, Faculty of Management, Gebze Technical University, 41400 Kocaeli, Turkey
*
Author to whom correspondence should be addressed.
Electronics 2023, 12(9), 1964; https://doi.org/10.3390/electronics12091964
Submission received: 4 December 2022 / Revised: 25 December 2022 / Accepted: 13 January 2023 / Published: 23 April 2023
(This article belongs to the Special Issue Artificial Intelligence Solutions and Applications for COVID-19 Pandemic)
Browse Figures
Versions Notes

Abstract

:
The purpose of this study was to prove the use of content and sentiment analysis to understand public discourse on Nytimes.com around the coronavirus (2019-nCOV) pandemic. We examined the pandemic discourses in the article contents, news, expert opinions, and statements of official institutions with natural language processing methods. We analyzed how the mainstream media (Nytimes.com) sets the community agenda. As a method, the textual data for the research were collected with the Orange3 software text-mining tool via the Nytimes.com API, and content analysis was conducted with Leximancer software. The research data were divided into three categories (first, mid, and last) based on the date ranges determined during the pandemic. Using Leximancer concept maps tools, we explained concepts and their relationships by visualizing them to show pandemic discourse. We used VADER sentiment analysis to analyze the pandemic discourse. The results gave us the distance and proximity positions of themes related to Nytimes.com pandemic discourse, revealed according to their conceptual definitions. Additionally, we compared the performance of six machine learning algorithms on the task of text classification. Considering the findings, it is possible to conclude that in Nytimes.com (2019-nCOV) discourse, some concepts have changed on a regular basis while others have remained constant. The pandemic discourse focused on specific concepts that were seen to guide human behavior and presented content that may cause anxiety to readers of Nytimes.com. The results of the sentiment analysis supported these findings. Another result was that the findings showed us that the contents of the coronavirus (2019-nCOV) articles supported official policies. It can be concluded that regarding the coronavirus (2019-nCOV), which has caused profound societal changes and has results such as death, restrictions, and mask use, the discourse did not go beyond a total of 15 main themes and about 100 concepts. The content analysis of Nytimes.com reveals that it has behavioral effects, such as causing fear and anxiety in people. Considering the media dependency of society, this result is important. It can be said that the agenda-setting of society does not go beyond the traditional discourse due to the tendency of individuals to use newspapers and news websites to obtain information.

1. Introduction

The coronavirus (2019-nCOV) pandemic, which is thought to have started in December 2019, has affected the global population’s daily life, lifestyle, and behavior as it spread across the world [1]. The pandemic has had such an impact that it became the number one problem on the agendas of all countries [2]. All newspapers, magazines, online and social media news, statements, and articles about the pandemic were used as sources of information. According to the reports on Forbes.com, pandemic-related news has been especially prevalent in the social media environment. Looking at 28 February 2020, it was estimated that 6.7 million people talked about the coronavirus (2019-nCOV) in one day, and as a result of this widespread discussion, people became worried [3]. Following this, people have relied on perceived high-quality news media to learn about the coronavirus (2019-nCOV) [4]. The coronavirus (2019-nCOV) crisis has led to significant increases in the consumption of news in digital media since the beginning of 2020, and people have felt the need to be extremely informed [5]. In conjunction with this, news reading behaviors have rapidly increased since the onset of the pandemic [6]. Moreover, it has been determined that people who turn to alternative media for news also use the mainstream media [7]. Nytimes.com, which is a part of the mainstream media, has shared articles, news stories, and instant data about the coronavirus (2019-nCOV) on its online website during the pandemic. In addition to the daily case status of the virus, expert comments, general evaluations, and articles were provided under many categories, such as sectoral effects. Therefore, it is important to analyze the contents of the Nytimes.com coronavirus (2019-nCOV) -related articles and reveal the relationships between emerging concepts. By analyzing the Nytimes.com unstructured textual data with artificial intelligence (AI) methods, how mainstream media pandemic discourse has evolved can be explored.
The methods used in the analysis of article content are transformed by digitalization, particularly with the use of AI-based software. The findings can be quickly obtained, and the human factors can be minimized. AI software interfaces are used for these processes. With natural language processing (NLP), one of the sub-branches of AI, keywords related to the research topic are queried, and the content and metadata of the articles published with these queried keywords are converted into a dataset. These resulting articles are converted into datasets that are automatically classified by going through various text processes. As a result of this classification, various findings are retrieved. In addition, by creating analytical models suitable for the research problem, the data are instantly evaluated, and the results are visualized. These processes are short, unlike traditional methods. It provides important information regarding textual data [8]. The most important contribution of the results of the analysis performed with NLP is that it reveals the relationships, which can provide further understanding to researchers. It also helps to discover new insights and open the “black box”. We used NLP because supervised and unsupervised learning, and particularly deep learning, are now widely used for modeling human language, but there is still a need for syntactic and semantic understanding, as well as domain expertise. These aspects are not necessarily present in the machine learning approaches that are currently in use. NLP is significant because it helps resolve ambiguity in language and offers a helpful quantitative structure to the data for applications that come after it, such as voice recognition or text analytics. Both benefits are vital for downstream applications. In this study, a similar method was used to analyze the textual contents of the coronavirus (2019-nCOV)-related articles on Nytimes.com. First, the descriptive statistics of the textual dataset were defined and sentiment measurements of the published content were conducted. Then, using Leximancer concept maps [9], we explained concepts and their relationships by visualizing them to identify pandemic discourse.
In conclusion, we believe that our work makes four primary contributions to the field, as follows:
  • We gathered information about how the mainstream media (Nytimes.com) sets the agenda for the community.
  • To evaluate the debate around the pandemic, we used VADER’s sentiment analysis. The findings provided us with the distance and closeness positions of a few topics connected to the pandemic discourse on Nytimes.com, disclosed in accordance with the conceptual definitions of those themes.
  • Using our dataset, we performed text classification, for which we created four alternative machine learning models. In these analysis results, the algorithm that demonstrated the best performance has been determined.
  • Concerning crisis communication monitoring, we have presented an overview of the emotions evoked and topics covered by Nytimes.com’s pandemic-related (2019-nCOV) coverage, which should be useful.
The remainder of this article is organized as follows: The literature related to our research is examined in Section 2, the proposed research method and data are reviewed in Section 3, and the results and evaluations of our study are examined in Section 4. The discussion and implications are stated in Section 5, and finally, the conclusions are presented in Section 6.

2. Related Works

Coronavirus (2019-nCOV) has been the most important agenda item for society and individuals in recent years. Studies have examined news in the mainstream media to understand and analyze the impact of coronavirus (2019-nCOV). The influence of national governments on news sources during the pandemic and the impact of the coronavirus (2019-nCOV) news on the tourism industry were discussed [10]. Exposure to news of coronavirus (2019-nCOV) in the mainstream media [11], as well as the effects of its news on consumer behavior and consumption habits [12], were examined. Van et al. evaluated the results of the use of coronavirus (2019-nCOV) in the news media [13]. Studies have been conducted to investigate the sectoral and social effects of the news media. These studies have achieved important results over the pandemic period. Research that analyzes public discourse during the pandemic often focuses on social media discourse [14,15,16,17,18,19]. These studies were combined with behavioral models and analytical frameworks to analyze public priorities and concerns [20].

2.1. Theoretical-Related Works

As in the agenda-setting theory [21], the dominant group, which has the power to control the mass media, also determines what people will discuss. Thus, the community agenda is formed [22]. Media giants publish news that is in line with their own worldviews and do not consider other views as important. The news is presented to the audience in a certain order. People tend to consider the news presented first as important, so the information given first always sets the agenda of people and society [23]. Developments in the theory also reveal new directions in research [22]. During the pandemic period, studies were conducted based on this theory, especially analyzing the coronavirus (2019-nCOV) discourses [24,25,26,27] of social media platforms. The Nytimes.com pandemic discourse will contribute to these findings because Nytimes.com has a prominent place in setting the agenda of society. Popular daily publications such as The New York Times and The Washington Post are “agenda setters” within the United States Media. These broadcasts then have a direct impact on local newspapers and television networks [28].

2.2. Practical-Related Works

Significant developments in the field of artificial intelligence (AI) have enabled these analyses to be performed. The ability to easily obtain data with various AI software applications and purposefully analyze them with NLP has led to the automation of research methods and a significantly shorter timeline for processes. The development of these methods has supplied the opportunity to perform content analysis used in qualitative research in a reliable way. Texts converted into numerical data with natural language processing have created the opportunity to make predictions and inferences using machine learning (ML) algorithms. NLP, a subfield of artificial intelligence (AI), is used in the analysis of unstructured textual data [29]. Software interfaces, such as Leximancer [30] and Orange3 [31], are used to perform NLP operations. Using these applications, large amounts of qualitative and unstructured data, consisting of text, are automatically classified, analyzed, and significant insights are identified. In this study, a dataset was created by collecting coronavirus (2019-nCOV)-related articles published on the Nytimes.com website. Leximancer, an AI-based NLP program, was used for content and sentiment analysis by passing the data through the steps of the working model.
Leximancer detects key concepts (or words) within blocks of text in the complete dataset based on their similarity and association with other words. The glossary generated and classified automatically for each set of data was created using machine learning. Classified texts are indexed as main themes, concepts, and compound concepts. Leximancer is also a computational content analysis tool for qualitative data analysis that calculates the frequency of concepts, their associations, and their proximity to each other [9,32]. The conceptual structure of articles in this study’s dataset was synthesized, indexed, measured, and visualized as in earlier studies of a similar nature. It has been used in many academic studies [33,34,35,36]. In this study, Leximancer 5.0 was used to analyze “coronavirus” articles on Nytimes.com.
Sentiment analysis is a classification method used in the perceptual analysis of unstructured textual contents [37]. There are various analytical methods for NLP. The most successful among these methods is the valence-aware dictionary for sentiment reasoning (VADER), which is available in the Natural Language Toolkit (NLTK) library. VADER is sensitive to both the polarity (positive or negative) and the intensity (strength) of perception [38]. VADER is directly applied to unlabeled text data [39]. The sentiment classification (positive or negative) of the text was performed by directly applying the VADER that was coded in Python.
There have been numerous studies analyzing pandemic events on social media using machine learning techniques and natural language processing [40]. One study used five machine learning algorithms (decision tree, logistic regression, k-nearest neighbors, random forest, and support vector machine) based on a historically tagged dataset of coronavirus (2019-nCOV) tweets to build a sensitivity classifier [41]. Twitter offers COVIDSensing, a real-time tool where they use topic modeling and sentiment analysis to analyze socio-economic issues related to coronavirus (2019-nCOV) on Really Simple Syndication and Telegram [42]. One study detected “flu” on Twitter with a support vector machine to classify negative and positive flu tweets [43]. Andreadis et al. [44] analyzed tweets about coronavirus (2019-nCOV) in Italy. They used logistic regression and random forests to classify fake news or misinformation. Apart from tweets, the authors of [45] combined random forest, stochastic gradient descent, and a logistic regression decision and created a predictive model for the retweetability of tweets posted regarding coronavirus (2019-nCOV).
In one study, a novel mesoscopic text representation was proposed to understand what happens at the mid-sentence level [46]. Additionally, they defined and introduced a network-based representation of textual data that separates the synchronic properties of a text from its diachronic properties. Mansoor et al. examined the sentiment of tweets related to coronavirus (2019-nCOV) and how sentiment in different countries has changed over time. They proposed a framework to monitor and analyze sentiment towards the novel coronavirus (2019-nCOV) over time on Twitter and sentiment towards Work from Home and Online Learning [47]. Another study introduced a new method of training word embeddings with information about context to induce word senses in the word sense induction task. Results show that word embeddings can be used to address limitations in a manually annotated corpus [48].
The above analyses focused on Twitter data. There are no studies analyzing Nytimes.com data. In our study, a sentiment measurement was performed using the VADER method. The data were then evaluated with ML models (random forest, naïve Bayes, support vector machine, multilayer perceptron, Bert, and Electra).

3. Data and Methods

We registered for a New York Times Developer Account membership to collect research data. Using the Access Token key given by the Nytimes.com API, we searched for “coronavirus” as a keyword for three different periods (first, mid, and last) with the Orange3 Text Data Mining [31] feature, as seen in Figure 1. For each query, we obtained 1000 text records. A total of 3000 text items were generated for the three queries. We converted this text data into a dataset and saved it in .txt and .csv formats separately for each period.
The contents of our dataset were as follows: We chose a time close to March 2020, the day the WHO declared the first case. First-term content ranged from 08 April to 16 June 2020, mid-term content from 26 February to 28 May 2021, and last-term content covered publications from 02 November to 28 December 2021. We chose the date ranges of the data by paying attention to the course of the pandemic, as shown in Figure 1. Each period holds approximately three months of text data. We removed duplicate data that could have influenced the analysis results. In this way, we have removed 102 rows of data from the analysis. In Figure 1, the datasets are sized as follows: first term (n = 950), mid-term (n = 975), and last term (n = 973).
For the descriptive analysis, we recorded the data separately for each period, merged the data into a single dataset (All.csv, n = 2898), and obtained the analysis results, as shown in Figure 2 for the Nytimes.com text dataset descriptive analysis. We then subjected this dataset to sentiment analysis. Finally, we created a project using the trial version of Leximancer software and performed a content analysis for our three-term data, separately.

4. Results

For descriptive analysis, we combined the data in a single dataset (All.csv, n = 2898), which was recorded separately for each period. According to the textual dataset of the three periods, articles were mostly in the “World” section, and the least in the “Books” section. The textual dataset’s article types were as follows: news (n = 2297), op-ed (n = 249), briefing (n = 114), interactive feature (n = 96), video (n = 59), obituary (n = 54), review (n = 10), editorial (n = 10), and news analysis (n = 9), (x2: 24,000.00, p = 0.0001, dof = 64).
The distribution of article types and section variables is shown in Figure 2 (x2: 7830.46, p = 0.0001, dof = 304).

4.1. Sentiment Analysis and Results

We measured the sentiment scores using VADER, which is the most suitable for our dataset and was coded using Python. VADER is a lexicon and rule-based sentiment analysis tool that is specifically attuned to sentiments expressed in social media. A sentiment lexicon is a list of lexical features (e.g., words) which are generally labeled according to their semantic orientation as either positive or negative. VADER has been found to be quite successful when dealing with social media texts, Nytimes editorials, movie reviews, and product reviews. This is because VADER not only obtains the positivity and negativity scores but also explains how positive or negative a sentiment is. Additionally, VADER sentiment analysis relies on a dictionary that maps lexical features to emotion intensities, known as sentiment scores. The sentiment score of a text can be obtained by summing up the intensity of each word in the text [38,49].
In comparison to more conventional approaches to sentiment analysis, VADER offers a number of benefits, including the following:
  • It performs very well on text, similar to that seen on social networking platforms, while easily generalizing to a variety of other fields.
  • It is developed using a generalizable, valence-based, human-curated gold standard sentiment lexicon, yet it does not need any training data.
  • It has a speed that allows it to be utilized online with streaming data, and it does not suffer from a speed–performance tradeoff to a significant degree.
Besides these advantages, VADER has some disadvantages, such as:
  • Analysis is language-specific.
  • Discriminating jargon, nomenclature, memes, or turns of phrase may not be recognized.
Table 1 shows that 1157 of the 2898 textual contents related to the coronavirus (2019-nCOV) on Nytimes.com were negative sentiments (40%), and 1741 were positive sentiments (60%). These findings of negative content can be interpreted as being quite high compared to the normal agenda.

4.2. Evaluation

In the supervised sentiment analysis, we used ML models and transformer-based approaches. The reason we chose these models was to compare the performance of traditional machine learning algorithms and transformer-based approaches.
We tested a variety of ML models and examined how well they performed relative to the performance of our trained model utilizing the text model as the baseline. The assessment was conducted in accordance with the usual classification metrics, which included precision, recall, F1-score, and accuracy. The evaluation results obtained from naive Bayes, random forest, support vector machine, and multilayer perceptron, as well as transformer-based approaches Bert and Electra, classification models that were trained using our data and text embeddings, respectively, are shown in Table 2 and Table 3, respectively.
According to the findings of the evaluation, our model performed better than the second model for each ML model (the accuracy of both was the same for naive Bayes and multilayer perceptron, the only ML models for which this was not the case, and transformer-based approaches Bert and Electra). When comparing ML models that were employed in the supervised task of sentiment analysis, multilayer perceptron earned the greatest scores of all evaluation metrics in both representation models. This was the case regardless of which model was chosen. Specifically, the F1-score for the first model was 0.73, while the F1-score for the second model was 0.71. Despite having higher performance values than the general average, the transformer-based approaches Bert and Electra were unable to achieve the highest model scores. To determine which algorithm produces the best results, class-wise performance measurements were performed across positive and negative classes. Table 4 shows the class-wise performance of the ML models and transformer-based approaches. Although the number of favorable comments was much smaller than the number of comments in other classes, all the trained models have the potential to perform more effectively with a dataset that is more evenly distributed.
Electra, which is one of the transformer-based techniques, earned the maximum performance for the negative class, while multilayer perceptron achieved the best performance for the positive class (87.3%).

4.3. Leximancer Content Analysis and Results

When the Leximancer analysis process was mapped, concepts were clustered into higher-level “themes”. Concepts that often appear together in the same text strongly attract one another. Therefore, they tend to be located close to each other in the map area and are shown as colored circles on the map [9]. The size, intersection, distance, proximity, and relationship of the colored circles give us an idea of the concepts. The concept map also includes the names of the main concepts in the text. These are shown on the map as gray labels (Figure 3, Figure 4 and Figure 5). Thus, the concepts created by the keywords in the analyzed texts and the themes created by the concepts help interpret the analyzed text and answer the research questions.
We adjusted Leximancer so that the first 33% of concepts appeared on the map. We analyzed and visualized the text content published on Nytimes.com over three periods. Thus, the discourse concepts and their relationships were proven.

4.3.1. Leximancer First-Term Content (08 April–16 June 2020) Analysis Results

Looking at the first-term published content (08 April–16 June 2020) in Figure 3, the “health” and “coronavirus” themes emerged. These themes have the concepts of “country, officials, United States, states, public, spread, announced, outbreak, government, infections, and million”, related to the central government. According to this result, it could be said that Nytimes.com brought official statements to the fore in its statements about coronavirus (2019-nCOV). In particular, the relationship between “officials”, “outbreaks”, “infections”, “spread”, and “millions” is remarkable. Regarding “spread”, “restrictions”, “thousands”, “outbreak”, and “economics”, contents were collected and related to the coronavirus. “President Trump” and “Covid” themes were classified close to each other. Content related to hospitals, care, and news has been connected with coronavirus (2019-nCOV). “Travel”, “latest”, and “died” were positioned a bit further from the center as they were related to coronavirus (2019-nCOV) but different in terms of concept type. This, however, was directly related. The sequential relation or link “coronavirus” ---> “pandemic” ---> “people” ---> “died” could be considered a summary of the first period. The effects of these components could be evaluated in terms of fear, anxiety, and panic.

4.3.2. Leximancer Mid-Term Content (26 February–28 May 2021) Analysis Results

The content of the textual data on the Nytimes.com website and the broadcasts made eight months after the first-term content were analyzed in Figure 4; this time, the “coronavirus” theme was in first place. Themes also differed in their content compared to the first period. The “vaccinated” and “vaccine” themes came to the fore as the entire content of the articles. The terms belonging to government institutions were collected in the content of “vaccinated” themes. “Vaccine” and “shot” themes were devised side-by-side in this term. The relationship between “Johnson and blood” in the content of the “vaccine” theme is remarkable. Moreover, in this period, the themes of “school”, “students”, “patients”, “family”, and “office” occupied a certain place in the content. Articles on the economy, travel restrictions, and deaths continued throughout the period. The emphasis on “pandemics”, “millions”, “cases”, “series”, “deaths”, “days”, “weeks”, “months”, and “years” show the negative reflections of this period. For the first time, the United States (“United States”) appeared with a different “India” concept. In addition to all these implications during this period, it shows that central institutions occupied a key place in the content. The “Centers for Disease Control” appeared as a new concept. The concepts of “pandemic” and “life” had been associated with the reopening of schools, and the content related to families. Time concepts, like any other, had a wide range of contents. Despite the “summer” season, the coronavirus (2019-nCOV) continued its effect in this period.

4.3.3. Leximancer Last-Term Content (2 November–28 December 2021) Analysis Results

In the last-term content, the most frequently used concept in the contents of the discourses was “coronavirus”, as shown in Figure 5. It was still in a central position compared to other themes and has gradually grown in volume. In this period, it was seen that the “variant” theme came to the fore and included “Omicron, cases, country, surge, restrictions, travel, spread, and infections” as concepts. The Omicron coronavirus variant (2019-nCOV) and its relationships with the contents of “cases, countries, surges, waves, restrictions, travel, spread, and infections” were remarkable. The relationship between “mandate”, “workers”, “vaccination”, “employees”, “workers”, and “home” contents has become clear. Contrary to other periods, “workers” became a theme during this period, and the discourse on mandatory vaccination of employees was included. Considering the width of the recently formed circle, we can conclude that employees have a key place in the content of discourse. “Biden, Administration, and Washington” concepts took place in the circle of “vaccine” themes. This showed that the US Administration had recently changed. Unlike President Trump, “Biden” has remained a concept lately—he could not form a circle. This showed that Trump was reported more often than Biden. Another important finding, unlike other periods, was that it created an “inflation” theme. In this theme circle, “price” became a concept. From this result, it can be concluded that curfews and restrictions started to manifest themselves as inflation during the pandemic period. Moreover, the concepts of “London” and “restriction” came to the fore in the “unvaccinated” theme. Restrictions were on the agenda for those who were not vaccinated. When you look at the discourse about travel concepts, it created content like other periods, and it became associated with “Omicron” in the “variant” theme. Finally, in this period, the discourse about the vaccination of “children” began to take place as content, unlike the earlier periods.

4.3.4. Leximancer Content Analysis Results for Three Periods, Themes and Concepts

Three periods (first, mid, and last) are listed in Table 5, to summarize the Nytimes.com coronavirus (2019-nCOV) discourse between the analyzed periods of themes and concepts that appeared from the Leximancer analysis results. Thus, a comparison of the above visualized findings was provided to better understand the periodic change of discourse.

5. Discussion and Implications

Leximancer analysis results are listed in Table 4 for the three periods. The concepts of coronavirus (2019-nCOV), restrictions, spread, people, children, school, and travel were included in the Nytimes.com discourse of coronavirus (2019-nCOV). In addition, the theme of coronavirus (2019-nCOV), which took place in all periods, took the second place after the first period. There was no discourse about vaccination in the first term, and it was revealed that vaccination occupied a large place in discourses in other periods. However, the last term showed content in the form of vaccinated and unvaccinated. Concepts related to time (day, day, week, time, latest, year) took place in all periods. Concepts related to official institutions were used in the contents of the articles, as an institution (government, federal, Centers for Disease Control, administration, authorities), as a title (President), and as a place (Washington, country, countries) in discourses in every period. This result shows that Nytimes.com has published the statements of these officials. Another finding is that the content related to death was included in the first two periods. Although there has been no decrease in the number of deaths from the coronavirus (2019-nCOV), the concept of “death” has not taken place recently. From this, it can be concluded that virus-related deaths have become normalized. It can also be said that the feeling of death in the first two periods was the anchor of convincing people to get vaccinated.
From the findings related to the results in Table 4 and the studies in the literature on coronavirus (2019-nCOV), the established consensus is that the mainstream news media do not directly spread lies and publish official statements as endorsements [50]. Nytimes.com content analysis results confirmed that this journalism is close to official statements and confirms political decisions. Despite research and articles on the impact of restrictions on human behavior [51,52], the concept of “lockdown” only took place in the first-term discourse. Although the restrictions continued in other periods, the content did not take place in the mid-term and long-term periods. It could not pass the concepts of “vaccine and vaccinated” in terms of frequency of use. Jia and Lu claimed in some of their media columns that China’s handling of coronavirus (2019-nCOV) created negative public opinion against China [53]. However, in Nytimes.com articles, apart from the USA, only India was mentioned as a country name. China, which is thought to be the origin of the coronavirus (2019-nCOV) and has the densest population, was not mentioned on Nytimes.com.
In studies conducted during the initial period of the coronavirus (2019-nCOV) crisis, it has been concluded that the pandemic will have serious economic consequences worldwide and will affect every country [54], and that it has a noticeable effect on global economic growth [55]. As in the literature, the contents of Nytimes.com were treated as “economic” and “business” concepts in the first two periods. As in other recent studies [56,57], “inflation and price” emerged in this study. Even though the pandemic has negatively affected most areas, the concepts of “coronavirus (2019-nCOV)” and “travel” have come together in every period. We can say that the companies that provide services in this field are among the sectors that have suffered the most, suffering large financial losses due to travel restrictions. Contrary to this finding, there was no discourse related to the most profitable sectors during the pandemic. Furthermore, about 60 different variants have been detected and reported in the United Kingdom as of 13 December 2019 regarding the coronavirus (2019-nCOV) [58], but the concept of “variant” was only found on Nytimes.com in the last-term period, referred to as “Omicron”. During the H1N1 “Swine Flu” pandemic, a mandatory employee influenza vaccination policy was introduced for the 2009–2010 flu season. Mandatory influenza vaccination among healthcare workers (HCWs) is part of a mandatory seasonal flu vaccination program [59,60]. In particular, policies have been formed regarding the compulsory influenza vaccination of health workers at the Children’s Hospitals [61]. When we analyzed recent articles on Nytimes.com, articles related to the mandatory vaccination of duty employees were included in the coronavirus (2019-nCOV) crisis, as in past practices. Vaccine- and vaccination-related contents were not included in the first-term discourse but became a full agenda in the other two periods. In addition, although the World Health Organization approved the Pfizer-BioNTech, Oxford-AstraZeneca, Moderna, and Janssen vaccines for emergency use, only the Johnson vaccine came to the fore in the article content.

6. Conclusions

According to the findings of the Nytimes.com content analysis regarding coronavirus (2019-nCOV), it has had many behavioral effects, such as inducing fear and anxiety in people, and it can be concluded that the discourse around coronavirus (2019-nCOV), which has caused great changes and results such as death, restrictions, and mask use, did not go beyond a total of 15 main themes and about 100 concepts. The results of content and sentiment analysis supported this inference. Considering the media dependency of society, this result is important. It can be said that the agenda-setting of society does not go beyond the traditional discourse due to the tendency of individuals to use newspapers and news websites to obtain information. In Leximancer analysis results, it was seen that the news contents were conceptualized as “virus”, “spread”, “millions”, “people”, and “died”, and this is supported by official discourse. It can be concluded that related concepts are agenda-setting.
There are periodic differences in the coronavirus (2019-nCOV) pandemic discourse on Nytimes.com. As a result of the Leximancer analysis, themes and concepts were found to run periodically parallel to the course of the pandemic. It can be concluded that the influence on the economy and restrictions on businesses, which took place in the first period, resulted in inflation and price increases in the last period. While there was no discourse about vaccination in the first period, vaccination discourse took place extensively in the last two. While the discourse in the contents of the first period was more formal and restrictive, it was seen that the restrictions decreased with the reopening of schools in the last two periods, and the discourse on using masks and getting vaccinated was included in the mid-term period. In Nytimes.com coronavirus (2019-nCOV) discourses, partial closeness to the literature and the pandemic process showed that the discourse was in parallel with official policies. Nytimes.com public discourse can support the horse-racing journalism view, with journalism being too close to official statements and approving political decisions.
The theoretical contribution of this research was to show how society’s agenda was created during the pandemic by analyzing Nytimes.com text contents using NLP methods. It can help break the “black box” of mainstream media. This contribution is also important considering the ability of the media to manipulate individuals through behavioral patterns such as fear and anxiety. It can reveal the importance of discourse in restraining society and persuading various practices. On the other hand, as a practical contribution, in this study, four distinct and widely used ML models and transformer-based approaches were used. The reason we chose these models was to compare the performance of traditional machine learning algorithms and transformer-based approaches. At this point, the text data that have been gathered were separated into two distinct datasets. To determine which algorithm produced the best results, class-wise performance measurements were performed across positive and negative classes. Electra, which is one of the transformer-based techniques, showed the best performance for the negative class, while multilayer perceptron achieved the best performance for the positive class (87.3%).
Future studies, not limited to Nytimes.com, may include discourses from other mainstream media outlets. In addition, the conceptual relationships that emerge in the findings and concern other disciplines may be the subject of future research.
The limitations of the study are that the Nytimes.com API only supplies access to 1000 articles at a time, so the full content of the pandemic discourse could not be accessed. It is also limited in that we analyzed only one mainstream media outlet.
This research has proven the possibilities afforded by the convenience and usefulness of natural language processing methods, which can process a large amount of textual data and provide insights into the sentiment and topics of the observed texts. When it comes to the task of analyzing public opinion and attitudes toward a variety of coronavirus (2019-nCOV)-related themes, natural language processing can complement the achievements of traditional approaches that are used in research in the fields of the humanities and social sciences. In this way, natural language processing can complement the achievements of traditional approaches used in research. When it comes to the duty of monitoring online discussion surrounding the coronavirus (2019-nCOV) pandemic, we feel that our work contributes to the pursuit of the increasing social media research that is now being conducted.

Author Contributions

Conceptualization, S.T.; methodology, S.T.; software, S.T.; validation, S.T. and B.S.; formal analysis, B.S.; investigation, B.S. and Y.S.B.; resources, S.T. and Y.S.B.; data curation, S.T.; writing—original draft preparation, S.T. and Y.S.B.; writing—review and editing, S.T. and B.S. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Data used in this study can be obtained by contacting the corresponding author.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Caduff, C. What Went Wrong: Corona and the World after the Full Stop. Med. Anthropol. Q. 2020, 34, 467–487. [Google Scholar] [CrossRef]
  2. Cheval, S.; Mihai Adamescu, C.; Georgiadis, T.; Herrnegger, M.; Piticar, A.; Legates, D.R. Observed and Potential Impacts of the COVID-19 Pandemic on the Environment. IJERPH 2020, 17, 4140. [Google Scholar] [CrossRef]
  3. Wiederhold, B.K. Using Social Media to Our Advantage: Alleviating Anxiety During a Pandemic. Cyberpsychol. Behav. Soc. Netw. 2020, 23, 197–198. [Google Scholar] [CrossRef]
  4. Viehmann, C.; Ziegele, M.; Quiring, O. Communication, Cohesion, and Corona: The Impact of People’s Use of Different Information Sources on Their Sense of Societal Cohesion in Times of Crises. J. Stud. 2022, 23, 629–649. [Google Scholar] [CrossRef]
  5. Newman, N. Reuters Institute Digital News Report 2020; Reuters Institute for the study of Journalism: Oxford, England, 2020. [Google Scholar]
  6. Kim, S.J.; Wang, X.; Malthouse, E.C. Digital News Readership and Subscription in the United States during COVID-19: A Longitudinal Analysis of Clickstream and Subscription Data from a Local News Site. Digit. J. 2022, 10, 1015–1036. [Google Scholar] [CrossRef]
  7. Andersen, K.; Shehata, A.; Andersson, D. Alternative News Orientation and Trust in Mainstream Media: A Longitudinal Audience Perspective. Digit. J. 2021, 1–20. [Google Scholar] [CrossRef]
  8. Lucy, L.; Demszky, D.; Bromley, P.; Jurafsky, D. Content Analysis of Textbooks via Natural Language Processing: Findings on Gender, Race, and Ethnicity in Texas U.S. History Textbooks. AERA Open 2020, 6, 233285842094031. [Google Scholar] [CrossRef]
  9. Leximancer User Guide. 2022, p. 149. Available online: https://www.doc.leximancer.com/ (accessed on 20 March 2022).
  10. Chen, H.; Huang, X.; Li, Z. A Content Analysis of Chinese News Coverage on COVID-19 and Tourism. Curr. Issues Tour. 2022, 25, 198–205. [Google Scholar] [CrossRef]
  11. Olagoke, A.A.; Olagoke, O.O.; Hughes, A.M. Exposure to Coronavirus News on Mainstream Media: The Role of Risk Perceptions and Depression. Br. J. Health Psychol. 2020, 25, 865–874. [Google Scholar] [CrossRef]
  12. Cruz-Cárdenas, J.; Zabelina, E.; Guadalupe-Lanas, J.; Palacio-Fierro, A.; Ramos-Galarza, C. COVID-19, Consumer Behavior, Technology, and Society: A Literature Review and Bibliometric Analysis. Technol. Forecast. Soc. Change 2021, 173, 121179. [Google Scholar] [CrossRef]
  13. Van Aelst, P.; Toth, F.; Castro, L.; Štětka, V.; Vreese, C.d.; Aalberg, T.; Cardenal, A.S.; Corbu, N.; Esser, F.; Hopmann, D.N.; et al. Does a Crisis Change News Habits? A Comparative Study of the Effects of COVID-19 on News Media Use in 17 European Countries. Digit. J. 2021, 9, 1208–1238. [Google Scholar] [CrossRef]
  14. Kurten, S.; Beullens, K. #Coronavirus: Monitoring the Belgian Twitter Discourse on the Severe Acute Respiratory Syndrome Coronavirus 2 Pandemic. Cyberpsychol. Behav. Soc. Netw. 2021, 24, 117–122. [Google Scholar] [CrossRef]
  15. Ellerich-Groppe, N.; Pfaller, L.; Schweda, M. Young for Old—Old for Young? Ethical Perspectives on Intergenerational Solidarity and Responsibility in Public Discourses on COVID-19. Eur. J. Ageing 2021, 18, 159–171. [Google Scholar] [CrossRef]
  16. Ayalon, L.; Chasteen, A.; Diehl, M.; Levy, B.R.; Neupert, S.D.; Rothermund, K.; Tesch-Römer, C.; Wahl, H.-W. Aging in Times of the COVID-19 Pandemic: Avoiding Ageism and Fostering Intergenerational Solidarity. J. Gerontol. Ser. B 2021, 76, e49–e52. [Google Scholar] [CrossRef]
  17. Xiang, X.; Lu, X.; Halavanau, A.; Xue, J.; Sun, Y.; Lai, P.H.L.; Wu, Z. Modern Senicide in the Face of a Pandemic: An Examination of Public Discourse and Sentiment About Older Adults and COVID-19 Using Machine Learning. J. Gerontol. Ser. B 2021, 76, e190–e200. [Google Scholar] [CrossRef]
  18. Pascual-Ferrá, P.; Alperstein, N.; Barnett, D.J. Social Network Analysis of COVID-19 Public Discourse on Twitter: Implications for Risk Communication. Disaster Med. Public Health Prep. 2022, 16, 561–569. [Google Scholar] [CrossRef]
  19. Xue, J.; Chen, J.; Chen, C.; Zheng, C.; Li, S.; Zhu, T. Public Discourse and Sentiment during the COVID 19 Pandemic: Using Latent Dirichlet Allocation for Topic Modeling on Twitter. PLoS ONE 2020, 15, e0239441. [Google Scholar] [CrossRef]
  20. Habib, M.A.; Anik, M.A.H. Impacts of COVID-19 on Transport Modes and Mobility Behavior: Analysis of Public Discourse in Twitter. Transp. Res. Rec. 2021, 2, 036119812110299. [Google Scholar] [CrossRef]
  21. Agenda-Setting Theory. 2022. Available online: https://wikipedia (accessed on 20 March 2022).
  22. McCombs, M.E.; Shaw, D.L.; Weaver, D.H. New Directions in Agenda-Setting Theory and Research. Mass Commun. Soc. 2014, 17, 781–802. [Google Scholar] [CrossRef]
  23. Littlejohn, S.W.; Foss, K.A. Theories of Human Communication, 10th ed.; Waveland Press: Long Grove, IL, USA, 2010; ISBN 978-1-4786-0939-1. [Google Scholar]
  24. Dai, Y.; Li, Y.; Cheng, C.-Y.; Zhao, H.; Meng, T. Government-Led or Public-Led? Chinese Policy Agenda Setting during the COVID-19 Pandemic. J. Comp. Policy Anal. Res. Pract. 2021, 23, 157–175. [Google Scholar] [CrossRef]
  25. Meutia, I.F.; Sujadmiko, B.; Yulianti, D.; Putra, K.A.; Aini, S.N. The Agenda Setting Policy for Hajj and Umrah in Post Pandemic. In Proceedings of the 2nd International Indonesia Conference on Interdisciplinary Studies (IICIS 2021), Amsterdam, The Netherlands, 28 October 2021; pp. 32–37. [Google Scholar]
  26. Liu, K.; Geng, X.; Liu, X. The Application of Network Agenda Setting Model during the COVID-19 Pandemic Based on Latent Dirichlet Allocation Topic Modeling. Front. Psychol. 2022, 13, 954576. [Google Scholar] [CrossRef] [PubMed]
  27. Wang, Q. Using Social Media for Agenda Setting in Chinese Government’s Communications during the 2020 COVID-19 Pandemic. J. Commun. Inq. 2022, 46, 01968599221105099. [Google Scholar] [CrossRef]
  28. Wikipedia Agenda-Setting Theory-Wikipedia. Available online: https://en.wikipedia.org/wiki/Agenda-setting_theory (accessed on 23 November 2022).
  29. Manning, C.; Surdeanu, M.; Bauer, J.; Finkel, J.; Bethard, S.; McClosky, D. The Stanford CoreNLP Natural Language Processing Toolkit. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics: System Demonstrations, Baltimore, MD, USA, 23–24 June 2014; pp. 55–60. [Google Scholar]
  30. Leximancer. Available online: https://www.leximancer.com/ (accessed on 20 March 2022).
  31. Orange Data Mining-Text Mining. Available online: https://orangedatamining.com/workflows/Text-Mining/ (accessed on 20 March 2022).
  32. Kanasa, D.H. An Introduction to Leximancer. Available online: https://www.leximancer.com/ (accessed on 20 March 2022).
  33. Angus, D.; Rintel, S.; Wiles, J. Making Sense of Big Text: A Visual-First Approach for Analysing Text Data Using Leximancer and Discursis. Int. J. Soc. Res. Methodol. 2013, 16, 261–267. [Google Scholar] [CrossRef]
  34. Wilk, V.; Cripps, H.; Capatina, A.; Micu, A.; Micu, A.-E. The State of #digitalentrepreneurship: A Big Data Leximancer Analysis of Social Media Activity. Int. Entrep. Manag. J. 2021, 17, 1899–1916. [Google Scholar] [CrossRef]
  35. Wilk, V.; Soutar, G.N.; Harrigan, P. Tackling Social Media Data Analysis: Comparing and Contrasting QSR NVivo and Leximancer. QMR 2019, 22, 94–113. [Google Scholar] [CrossRef]
  36. Tunca, S.; Sezen, B.; Wilk, V. An Exploratory Content and Sentiment Analysis of The Guardian Metaverse Articles Using Leximancer and Natural Language Processing. Cyberpsychol. Behav. Soc. Netw. 2022, 26, 56–78. [Google Scholar]
  37. Pang, B.; Lee, L.; Vaithyanathan, S. Thumbs up? Sentiment Classification Using Machine Learning Techniques. arXiv 2002, arXiv:cs/0205070. [Google Scholar]
  38. Hutto, C.J.; Gilbert, E. VADER: A Parsimonious Rule-Based Model for Sentiment Analysis of Social Media Text. In Proceedings of the International AAAI Conference on Web and Social Media, Ann Arbor, MI, USA, 1–4 June 2014. [Google Scholar]
  39. Beri, A. Sentimental Analysis Using Vader. Interpretation and Classification of…|by Aditya Beri|Towards Data Science. Available online: https://towardsdatascience.com/sentimental-analysis-using-vader-a3415fef7664?gi=ee44b81a54cb (accessed on 3 February 2022).
  40. Qin, Z.; Ronchieri, E. Exploring Pandemics Events on Twitter by Using Sentiment Analysis and Topic Modelling. Appl. Sci. 2022, 12, 11924. [Google Scholar] [CrossRef]
  41. Zhang, X.; Saleh, H.; Younis, E.M.G.; Sahal, R.; Ali, A.A. Predicting Coronavirus Pandemic in Real-Time Using Machine Learning and Big Data Streaming System. Complexity 2020, 2020, 6688912. [Google Scholar] [CrossRef]
  42. Sepúlveda, A.; Periñán-Pascual, C.; Muñoz, A.; Martínez-España, R.; Hernández-Orallo, E.; Cecilia, J.M. COVIDSensing: Social Sensing Strategy for the Management of the COVID-19 Crisis. Electronics 2021, 10, 3157. [Google Scholar] [CrossRef]
  43. Aramaki, E.; Maskawa, S.; Morita, M. Twitter Catches The Flu: Detecting Influenza Pandemics Using Twitter. In Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, Edinburgh, Scotland, UK, 27–31 July 2011. [Google Scholar]
  44. Andreadis, S.; Antzoulatos, G.; Mavropoulos, T.; Giannakeris, P.; Tzionis, G.; Pantelidis, N.; Ioannidis, K.; Karakostas, A.; Gialampoukidis, I.; Vrochidis, S.; et al. A Social Media Analytics Platform Visualising the Spread of COVID-19 in Italy via Exploitation of Automatically Geotagged Tweets. Online Soc. Netw. Media 2021, 23, 100134. [Google Scholar] [CrossRef] [PubMed]
  45. Mahdikhani, M. Predicting the Popularity of Tweets by Analyzing Public Opinion and Emotions in Different Stages of COVID-19 Pandemic. Int. J. Inf. Manag. Data Insights 2022, 2, 100053. [Google Scholar] [CrossRef]
  46. de Arruda, H.F.; Silva, F.N.; Marinho, V.Q.; Amancio, D.R.; Costa, L. da F. Representation of Texts as Complex Networks: A Mesoscopic Approach. J. Complex Netw. 2018, 6, 125–144. [Google Scholar] [CrossRef]
  47. Mansoor, M.; Gurumurthy, K.; Prasad, V.R. Global Sentiment Analysis Of COVID-19 Tweets Over Time. arXiv 2020, arXiv:2010.14234. [Google Scholar]
  48. Corrêa, E.A.; Amancio, D.R. Word Sense Induction Using Word Embeddings and Community Detection in Complex Networks. Phys. A Stat. Mech. Its Appl. 2019, 523, 180–190. [Google Scholar] [CrossRef]
  49. Pandey, S. Simplifying Sentiment Analysis Using VADER in Python (on Social Media Text)|by Parul Pandey|Analytics Vidhya|Medium. Available online: https://medium.com/analytics-vidhya/simplifying-social-media-sentiment-analysis-using-vader-in-python-f9e6ec6fc52f (accessed on 25 December 2022).
  50. Quandt, T.; Boberg, S.; Schatto-Eckrodt, T.; Frischlich, L. Pandemic News: Facebook Pages of Mainstream News Media and the Coronavirus Crisis—A Computational Content Analysis. arXiv 2020, arXiv:2005.13290. [Google Scholar]
  51. Ahmed, S.; Khaium, M.O.; Tazmeem, F. COVID-19 Lockdown in India Triggers a Rapid Rise in Suicides Due to the Alcohol Withdrawal Symptoms: Evidence from Media Reports. Int. J. Soc. Psychiatry 2020, 66, 827–829. [Google Scholar] [CrossRef]
  52. Lemenager, T.; Neissner, M.; Koopmann, A.; Reinhard, I.; Georgiadou, E.; Müller, A.; Kiefer, F.; Hillemacher, T. COVID-19 Lockdown Restrictions and Online Media Consumption in Germany. IJERPH 2020, 18, 14. [Google Scholar] [CrossRef]
  53. Jia, W.; Lu, F. US Media’s Coverage of China’s Handling of COVID-19: Playing the Role of the Fourth Branch of Government or the Fourth Estate? Glob. Media China 2021, 6, 8–23. [Google Scholar] [CrossRef]
  54. Donthu, N.; Gustafsson, A. Effects of COVID-19 on Business and Research. J. Bus. Res. 2020, 117, 284–289. [Google Scholar] [CrossRef]
  55. Crs, R. Global Economic Effects of COVID-19; Congressional Research Service: Washington, DC, USA, 2020; Volume 84, pp. 20–115. [Google Scholar]
  56. Apergis, E.; Apergis, N. Inflation Expectations, Volatility and Covid-19: Evidence from the US Inflation Swap Rates. Appl. Econ. Lett. 2021, 28, 1327–1331. [Google Scholar] [CrossRef]
  57. Coluccia, B.; Agnusdei, G.P.; Miglietta, P.P.; De Leo, F. Effects of COVID-19 on the Italian Agri-Food Supply and Value Chains. Food Control 2021, 123, 107839. [Google Scholar] [CrossRef] [PubMed]
  58. Wise, J. COVID-19: New Coronavirus Variant Is Identified in UK. BMJ 2020, 371, m4857. [Google Scholar] [CrossRef] [PubMed]
  59. Babcock, H.M.; Gemeinhart, N.; Jones, M.; Dunagan, W.C.; Woeltje, K.F. Mandatory Influenza Vaccination of Health Care Workers: Translating Policy to Practice. Clin. Infect. Dis. 2010, 50, 459–464. [Google Scholar] [CrossRef]
  60. Hakim, H.; Gaur, A.H.; McCullers, J.A. Motivating Factors for High Rates of Influenza Vaccination among Healthcare Workers. Vaccine 2011, 29, 5963–5969. [Google Scholar] [CrossRef]
  61. Douville, L.E.; Myers, A.; Jackson, M.A.; Lantos, J.D. Health Care Worker Knowledge, Attitudes, and Beliefs Regarding Mandatory Influenza Vaccination. Arch. Pediatr. Adolesc. Med. 2010, 164, 33–37. [Google Scholar] [CrossRef]
Electronics 12 01964 g001 550
Figure 1. Structure and process of the Nytimes.com discourse analysis.
Figure 1. Structure and process of the Nytimes.com discourse analysis.
Electronics 12 01964 g001
Electronics 12 01964 g002 550
Figure 2. Nytimes.com article dataset box plot for descriptive analysis.
Figure 2. Nytimes.com article dataset box plot for descriptive analysis.
Electronics 12 01964 g002
Electronics 12 01964 g003 550
Figure 3. First-term concept map of the coronavirus (2019-nCOV) news, showing the nature of themes and concepts on Nytimes.com.
Figure 3. First-term concept map of the coronavirus (2019-nCOV) news, showing the nature of themes and concepts on Nytimes.com.
Electronics 12 01964 g003
Electronics 12 01964 g004 550
Figure 4. Mid-term concept map of the coronavirus (2019-nCOV) news, showing the nature of themes and concepts on Nytimes.com.
Figure 4. Mid-term concept map of the coronavirus (2019-nCOV) news, showing the nature of themes and concepts on Nytimes.com.
Electronics 12 01964 g004
Electronics 12 01964 g005 550
Figure 5. Last-term concept map of the coronavirus (2019-nCOV) news, showing the nature of themes and concepts on Nytimes.com.
Figure 5. Last-term concept map of the coronavirus (2019-nCOV) news, showing the nature of themes and concepts on Nytimes.com.
Electronics 12 01964 g005
Table
Table 1. Sentiment analysis results.
Table 1. Sentiment analysis results.
Row LabesCount of TextPercentage (%)
Positive174160%
Negative115740%
Total2898100%
Table
Table 2. Calculations of accuracy, recall, and F1-score as macro-averages for evaluations using first data embeddings.
Table 2. Calculations of accuracy, recall, and F1-score as macro-averages for evaluations using first data embeddings.
MethodPrecisionRecallF1-ScoreAccuracy
Random Forest0.640.720.640.78
Naïve Bayes0.580.600.590.86
Support Vector Machine0.590.610.600.85
Multilayer Perceptron0.730.740.730.86
Bert0.660.650.560.81
Electra0.700.700.620.84
Table
Table 3. Calculations of accuracy, recall, and F1-score as macro-averages for baseline evaluations using second data embeddings.
Table 3. Calculations of accuracy, recall, and F1-score as macro-averages for baseline evaluations using second data embeddings.
MethodPrecisionRecallF1-ScoreAccuracy
Random Forest0.620.700.620.76
Naïve Bayes0.560.580.570.84
Support Vector Machine0.570.590.580.83
Multilayer Perceptron0.710.720.710.84
Bert0.650.670.620.83
Electra0.690.680.590.82
Table
Table 4. Class-wise performance of the ML models and transformer-based approaches.
Table 4. Class-wise performance of the ML models and transformer-based approaches.
ClassRFNBSVMMPBertElectra
Positive74.586.185.787.384.681.2
Negative78.282.681.382.582.284.3
Table
Table 5. Leximancer content analysis results for the three periods, themes and concepts.
Table 5. Leximancer content analysis results for the three periods, themes and concepts.
First-Term Content
(8 April–16 June 2020)		Mid-Term Content
(26 February–28 May 2021)		Last-Term Content
(2 November–28 December 2021)
ThemesConceptsThemesConceptsThemesConcepts
healthcountry, officials, United States, states, public, spread, announced, outbreak, government, infections, millionCoronaviruspeople, United States, country, months, India, world, Washington, millions, virus, administrationCoronavirusvaccinated, announced, week, positive, testing, months, public
Coronaviruspandemic, people, cases, Washington, world, virusvaccinatedannounced, fully, Centers for Disease Control,
Americans, time, federal, public, masks, government, summer		variantOmicron, cases, country, surge, restrictions, travel, spread, infections
monthsmonth, March, city, lockdownvaccineCovid, health, officials, week, vaccines, weekspandemicpeople, United States, world, countries
homeyear, weeks, New YorkcasesJohnson, vaccination, doses, million,
number, rare		covidhealth, officials, government, virus
timeNew York City, down, workpandemicrestrictions, month, past, American, travelvaccinebooster, shots, federal, administration, Biden
dayweek, care, dayssummernight, economicworkersmandate, vaccination, employees
spreadoutbreak, businessesdiedothers, seriesdayChristmas
duringworkersschoolreopening, home,
students		downvaccines, Americans, rising, case
diedseries, othersyearduring, returnchildrenchildren, city, company
lockdownlockdownfamilyreopeningweekspolicy
CovidlongdaysnewsinflationPrices, time
businesseseconomic, restrictions, companyshottakedownnearly
President TrumpWashingtonstudentsstudentslongOmicron
traveltravelpatientshospitalunvaccinatedDuring, wave
latestlatestofficelifeschoolPercent
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Tunca, S.; Sezen, B.; Balcioglu, Y.S. Content and Sentiment Analysis of The New York Times Coronavirus (2019-nCOV) Articles with Natural Language Processing (NLP) and Leximancer. Electronics 2023, 12, 1964. https://doi.org/10.3390/electronics12091964

AMA Style

Tunca S, Sezen B, Balcioglu YS. Content and Sentiment Analysis of The New York Times Coronavirus (2019-nCOV) Articles with Natural Language Processing (NLP) and Leximancer. Electronics. 2023; 12(9):1964. https://doi.org/10.3390/electronics12091964

Chicago/Turabian Style

Tunca, Sezai, Bulent Sezen, and Yavuz Selim Balcioglu. 2023. "Content and Sentiment Analysis of The New York Times Coronavirus (2019-nCOV) Articles with Natural Language Processing (NLP) and Leximancer" Electronics 12, no. 9: 1964. https://doi.org/10.3390/electronics12091964

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop