The COVID-19 pandemic presented unprecedented challenges, necessitating an understanding of public opinion to inform public health strategies. While previous research has utilized social networking services (SNS) to analyze public sentiment, this approach has limitations. SNS data primarily reflects the opinions of younger demographics and may lack the depth and context offered by more established media. Furthermore, much of the existing research is short-term, failing to capture the evolution of public opinion over the prolonged duration of the pandemic. This study addresses these limitations by analyzing reader posts submitted to three major Japanese newspapers – Yomiuri, Mainichi, and Asahi – during four periods of state-of-emergency declarations in Japan. Japanese newspapers maintain high readership, especially amongst older demographics, and reader submissions often include personal details, offering a richer dataset than that available through anonymous SNS posts. Analyzing these submissions allows for a longitudinal study of public opinion during a sustained period of the pandemic, providing insights not readily available through short-term or SNS-based analyses. Given Japan's high newspaper readership and relatively stable social and infection control context throughout the pandemic, this offers a unique opportunity to study public opinion trends under these specific circumstances.

Existing research on public opinion during the COVID-19 pandemic has primarily relied on data from social networking services (SNS), questionnaires, and interviews. SNS studies have revealed various attitudes and emotions related to lockdowns, vaccines, and the pandemic in general. However, these studies often focus on short-term trends and primarily capture the perspectives of younger, digitally active populations, overlooking the opinions of older demographics. The present study addresses this gap by examining reader posts in traditional print newspapers, a medium more likely to reach a broader age range and offer more detailed and thoughtful contributions.

This study employed a document analysis of readers' posts submitted to the Yomiuri, Mainichi, and Asahi newspapers, three of the largest newspapers in Japan, during the four periods of emergency declarations from April 2020 to September 2021. A total of 1910 posts were collected. Latent Dirichlet Allocation (LDA), a topic modeling technique, was used to identify underlying themes within the text data. The optimal number of topics was determined using various evaluation metrics from the ldatuning package in R, resulting in the extraction of ten distinct topics. The top ten words for each topic were then examined to assign meaningful labels. The Jonckheere-Terpstra test was employed to examine trends in topic occurrences across the four emergency periods. Kendall rank correlation coefficients were calculated to assess the effect size of these trends. The study excluded the frequent keywords "COVID-19" and "Corona-Ka" from the topic analysis to avoid skewing the results. Statistical significance was determined using the Jonckheere-Terpstra test, and effect sizes were evaluated using Kendall rank correlation coefficients. Effect sizes of 0.1 or higher were considered meaningful.

The LDA analysis yielded ten topics categorized into three themes: Life, Awareness of the Emergency, and Policy. The "Life" theme included topics such as Family, Daily Life in the COVID-19 Disaster, Education in the COVID-19 Disaster, The Importance of Humanity, and Daily Life Unrelated to COVID-19. The "Awareness of the Emergency" theme comprised Awareness of Being a Party to an Emergency and Concerns about the Medical Environment. The "Policy" theme included Domestic and Foreign Policies, Opposition to Hosting the Tokyo Olympics, and Criticisms of the Japanese Government. Analysis revealed a statistically significant decrease in the "Awareness of Being a Party to an Emergency" topic across the four emergency periods (r = -0.189, *p* < 0.000). This finding suggests a decline in public concern regarding infection prevention measures and the overall emergency status, potentially indicative of pandemic fatigue. Other topics exhibited less pronounced trends, with some showing small increases or decreases, but none reached a statistically meaningful effect size (above 0.1). In short, while public awareness of the immediate emergency declined, other public concerns and opinions remained fairly consistent throughout the pandemic.

The findings highlight the value of analyzing newspaper reader posts for a comprehensive understanding of public opinion during a prolonged crisis. Unlike SNS data that often reflects more acute or rapidly evolving concerns, newspaper posts offer a glimpse into the enduring concerns and anxieties of the population. The decrease in "Awareness of Being a Party to an Emergency" likely reflects pandemic fatigue, a phenomenon characterized by a decline in motivation to follow recommended preventive measures. However, the stability of other topics, such as family, education, and concerns about the healthcare system, indicates that daily life concerns persisted throughout the pandemic. These sustained concerns warrant sustained attention from policymakers and public health officials. This study demonstrates the importance of considering both short-term and long-term trends in public opinion when developing and implementing public health interventions during prolonged crises. The stable presence of opinions related to daily life, even amidst the stress of the pandemic, underscores the need to address these broader societal concerns.

This study provides valuable insights into the evolution of public opinion in Japan during the prolonged COVID-19 pandemic. By analyzing reader posts from major Japanese newspapers, the study reveals the sustained presence of everyday concerns alongside a decline in acute awareness of the emergency situation. This underscores the need for a holistic approach to public health communication and policy-making, addressing both immediate and long-term concerns. Future research could explore the interplay of different media sources in shaping public opinion, investigate the influence of demographic factors on pandemic fatigue, and examine the impact of policy responses on public sentiment over time.

This study's findings are limited to the specific sample of reader posts submitted to three major Japanese newspapers. The selection process of published posts by newspaper editors may introduce bias. Therefore, the results may not be fully generalizable to the entire Japanese population or even all newspaper readers. Additionally, the LDA topic modeling is sensitive to the choice of parameters and the interpretation of extracted topics. Further research with larger and more diverse datasets is needed to strengthen the generalizability of the findings and refine the topic modeling process.