The COVID-19 pandemic had a devastating impact globally, particularly on health, the economy, and mental well-being. Strict social distancing measures led to widespread business closures and unprecedented unemployment rates, exacerbating existing mental health challenges. While sentiment analysis studies have examined public emotions on social media during the pandemic, this study uniquely focuses on the portrayal of unemployment in a mainstream news source, the *New York Times*, to understand how public perception was shaped by this important medium. The researchers aim to explore the emotional dynamics conveyed by the *New York Times* regarding pandemic-induced unemployment in 2020, how these emotions changed over time, and the significant events and themes linked to these emotional shifts. The study's significance lies in understanding the role of mainstream news media in shaping public perception and providing information during times of severe crisis, especially when contrasted with the often more individualistic expression found on social media.

Sentiment analysis, encompassing the study of emotions, opinions, and attitudes, employs various methods including supervised machine learning and lexicon-based approaches. The study notes a growing trend toward hybrid methods. Existing research has primarily analyzed sentiments on social media regarding the pandemic's economic impact and government responses. Studies have used Twitter data to compare macroeconomic responses across countries with varying income levels, analyze sentiments concerning the economy across demographics and geographies, and even nowcast unemployment rates. Other work has investigated public perception of government efforts to combat unemployment, finding both positive and negative sentiments depending on the specific measures and the platforms used. However, this research highlights a gap: the lack of sentiment analysis on mainstream news media concerning unemployment during the pandemic. Because of the differing roles of social media (individual-focused) and mainstream media (broader societal focus), studying the sentiment expressed in mainstream news is crucial for a complete understanding of public perception.

The study used a corpus of unemployment-related news articles from the *New York Times* in 2020, obtained through LexisNexis. Articles mentioning terms like "layoff," "lay off," "laid off," and "unemploy*" were included, resulting in a corpus of 4921 articles (8,016,170 words). Data cleaning involved lowercasing, tokenization, lemmatization, part-of-speech tagging, and removal of duplicates, URLs, punctuation, stop words, and non-English characters. Sentiment and emotion analysis utilized the NRC Emotion Lexicon, a lexicon-based approach identifying positive/negative sentiments and eight basic emotions (anger, anticipation, surprise, trust, disgust, fear, joy, sadness). Python scripts matched words in the corpus with associated emotions, with calculations performed monthly to track changes over time. Corpus-based discourse analysis, using techniques such as frequency analysis, collocate analysis, and cluster analysis (via AntConc), complemented the lexicon-based approach to identify themes and events related to the extracted emotions. This methodological triangulation aimed for a more comprehensive understanding of the emotional dynamics presented in the news articles. The researchers analyzed the correlation between news coverage and the actual unemployment rate, as well as the correlation between positive and negative sentiments, and explored the temporal trajectory of both sentiments and emotions throughout the year.

The number of unemployment-related articles in the *New York Times* strongly correlated with the actual unemployment rate (r=0.868, p<0.05). Sentiment analysis revealed significantly higher positive sentiment than negative sentiment throughout 2020, contradicting some social media studies that found predominantly negative sentiment. Positive and negative sentiments showed a significant negative correlation (r=-0.624, p<0.05), with fluctuations occurring around February, March, and August. Emotion analysis showed "trust" and "anticipation" as the most prominent positive emotions, while "fear" and "sadness" were the most prominent negative emotions. All eight emotions generally followed the same temporal pattern as the sentiments. Further analysis of words associated with "trust" and "anticipation" showed a strong focus on government economic policies and their expected impacts on daily life. Analysis of words associated with "fear" revealed connections to pandemic-related risks, economic concerns, and notably, the George Floyd protests in June 2020, highlighting the intersection of unemployment and racial inequality. The increase of "sadness" correlated strongly with pandemic-related deaths and economic hardships.

The findings suggest that the *New York Times*' representation of unemployment in 2020 was influenced by multiple factors: the pandemic's trajectory, government policy responses, and the racial justice protests. The initial surge in negative sentiment aligns with the pandemic's impact and rising unemployment; however, the subsequent shift towards more positive sentiment may reflect the government's policy response, along with a reduction in deaths and improved treatment options. The significant role of the George Floyd protests and their amplification through the news highlights the intertwined nature of economic hardship and social justice concerns. The contrast between these findings and the predominantly negative sentiment detected in social media analyses points to fundamental differences in the functions and scopes of social media and mainstream news. Mainstream news media, aiming for objectivity, may emphasize the government's response, thereby contributing to the higher positive sentiment observed.

This study offers valuable insights into how mainstream news media framed public perception of unemployment during the pandemic. The dominance of positive sentiment highlights the potential of solution-focused reporting to mitigate negative emotional responses during a crisis. The study's limitations include the reliance on a single news source and the use of a lexicon-based approach that may not fully capture the nuances of language. Future research could expand the dataset to include multiple news outlets and incorporate machine learning techniques for a more comprehensive analysis.

The study's reliance on a single newspaper (*New York Times*) limits the generalizability of the findings. The lexicon-based sentiment analysis, while complemented by discourse analysis, may not capture all contextual nuances of language use. The study primarily focuses on the emotional representation of unemployment, not necessarily on its actual effect on people's mental health. Further research could address these limitations by expanding the data sources and methodology.