2.1 Ethics statement

The present study utilized publicly available, anonymized data from the “Sometrend” big data analytics platform. The research did not involve direct interaction with human subjects, and no personally identifiable information (PII) was collected, processed, or stored. Since the study relied exclusively on existing, publicly accessible data in a manner that does not allow for the identification of individuals, it was determined to be exempt from formal Institutional Review Board (IRB) approval under the applicable ethical guidelines for secondary data analysis.

2.2 Data collection

To empirically investigate the sentiments and key issues associated with emergency departments (EDs) in South Korea during the post-pandemic period, this study utilized Sometrend, an AI-based big data analytics platform specialized for processing unstructured text data. Sometrend aggregates data from major search engines and social media channels in Korea, enabling the extraction of keywords, sentiment polarity, and semantic networks from vast amounts of online discourse.

Data collection was conducted for the period from January 1, 2024, to December 31, 2024, capturing the height of the medical void crisis. The search keyword was “emergency room,” and data were harvested from diverse digital channels including blogs, news articles, online communities, and social media platforms such as Instagram and YouTube. A total of 1,359,146 data points were collected for analysis.

2.3 Data analysis

The collected unstructured text data were analyzed using text mining techniques, following a two-stage process. Before sentiment classification, the raw text data underwent a rigorous preprocessing pipeline. This included: (1) noise reduction by removing HTML tags, special characters, and URLs; (2) tokenization using a Korean morphological analyzer to identify root words; and (3) deduplication to ensure that identical posts shared across multiple platforms did not skew the frequency counts. This pipeline ensured the purity of the unstructured data before it was fed into the AI engine.

Stage 1: Sentiment Analysis First, a sentiment analysis was performed to quantify healthcare consumers’ emotional responses to ED services. Sentiment analysis classifies text data into positive, negative, or neutral categories, allowing researchers to objectively measure public opinion and satisfaction [11]. With the proliferation of social network services (SNS), patients increasingly express their unfiltered emotions online, making sentiment analysis a critical tool for understanding patient experiences [12]. In this study, we specifically focused on identifying the proportion of negative sentiments to diagnose the current state of ED services in South Korea.

Stage 2: Keyword Frequency and Association Analysis Second, we conducted keyword frequency and association analyses to identify the specific causes of negative sentiment. Keyword frequency analysis extracts the most recurrent terms to determine core themes within a dataset [13], effectively revealing trends and shifts in public interest [14]. Furthermore, association analysis examines the co-occurrence of words to uncover the contextual relationships between key concepts [15]. This approach has been widely adopted in healthcare research to map knowledge structures and track emerging trends [16]. The sentiment analysis was conducted using Sometrend’s hybrid engine, which combines a pre-defined sentiment lexicon with machine learning algorithms specialized for the Korean language. To ensure the reliability of the automated classification, two independent researchers manually validated a random sample of 1,000 data points, achieving an inter-rater reliability (Cohen’s Kappa) of 0.84. Furthermore, all data collection complied with the terms and conditions of the source platforms via Sometrend’s API.

3.1 Data collection overview

The data collection process and summary are presented in Table 1. To investigate sentiments regarding emergency departments (EDs) in South Korea, data were aggregated using the Sometrend platform for the period from January 1, 2024, to December 31, 2024. Data were harvested from diverse digital channels, including major search engines (Naver, Daum), social media (Twitter/X, Instagram, YouTube), blogs, news outlets, and online communities. A total of 1,359,146 data points were collected and utilized for the final analysis

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Summary of data collection.

https://doi.org/10.1371/journal.pone.0352142.t001

3.2 Sentiment analysis

The distribution of sentiments regarding EDs in South Korea during the study period is illustrated in Fig 1. The results revealed that negative sentiment was the most prevalent, accounting for 46% of the total data. Positive sentiment constituted 32%, while neutral sentiment accounted for 22%. This predominance of negative sentiment indicates a high level of concern or dissatisfaction among healthcare consumers regarding ED services during the healthcare crisis. A Chi-square (χ2) test was performed to determine the statistical significance of these sentiment proportions. The analysis confirmed a significant difference (χ2 = 117,288.2, df = 2, p < .001), proving that negative sentiment was disproportionately dominant during the study period. To address potential platform-specific differences, an additional inferential statistical analysis (Chi-square test of independence) was conducted to compare sentiment distributions between News sources and Social Media. The analysis revealed a significant association between the source type and sentiment polarity (χ2 = 257.81, df = 2, p < .001). Specifically, News content displayed a significantly higher concentration of negative framing (48.2%) regarding the ‘medical void’ compared to user-generated content on Social Media (46.4%), suggesting that media coverage played a key role in amplifying public concern (Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Sentiment analysis results.

Sentiment categories were defined based on Sometrend’s standardized emotional lexicon for the Korean language. Overall sentiment distribution regarding EDs: Negative (45.8%), Positive (32.4%), and Neutral (21.9%). Goodness-of-fit test results: χ2 = 117,288.2, df = 2, p < .001. Sentiment categories were defined based on Sometrend’s standardized emotional lexicon.

https://doi.org/10.1371/journal.pone.0352142.g001

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Comparison of sentiment distribution between news and social media.

https://doi.org/10.1371/journal.pone.0352142.t002

3.3 Keyword frequency and association analysis

To identify the specific drivers of negative sentiment, the top keywords and their associated terms were analyzed. As shown in Table 3, the most frequent negative keyword was “Pain” (26,596), followed by “Overtreatment” (21,394), “Emergency Situation” (14,879), “Waiting Time” (14,410), and “Medical Void” (13,521). Notably, terms reflecting the 2024 healthcare crisis in Korea, such as “Emergency Room Runarounds” (10,363) and “Medical Personnel Shortage” (9,655), also ranked highly. The associated words for these keywords included descriptors such as “painful,” “unnecessary,” “sudden,” “stressful,” and “concerned,” which highlight the physical and psychological distress experienced by patients during this period.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Top 10 negative keywords and associated words (raw frequency counts).

https://doi.org/10.1371/journal.pone.0352142.t003

Furthermore, the analysis of keywords categorized by time, place, and situation provided contextual insights into ED utilization patterns, as summarized in Table 4. Regarding temporal factors, ED-related searches were most frequent during “Holidays” (39,811), followed by “Nighttime” (33,000) and “Weekends” (28,069), reflecting an increased reliance on emergency services when primary care clinics are typically closed. In terms of location, “Seoul” (35,055) was the most frequently mentioned, followed by “University Hospitals” (30,898) and “Pharmacies” (29,090). Finally, the most common situational contexts were “Emergency Situation” (25,055) and “Death” (17,703), underscoring the critical and often life-threatening nature of ED visits during the study period.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Top keywords by time, place, and situation (raw frequency counts).

https://doi.org/10.1371/journal.pone.0352142.t004

4.1 The prevalence of negative sentiment in the context of a “medical void”

This study empirically identified that negative sentiment (46%) was the dominant emotional response among South Korean healthcare consumers regarding emergency departments (EDs) in the post-pandemic era. This high level of dissatisfaction is significant when compared to general ED satisfaction studies. While EDs inherently involve high-stress situations, the prevalence of keywords such as “medical void” and “emergency room runarounds” suggests that the current negativity is not merely due to clinical outcomes but stems from systemic failures in access. However, these results regarding a “trust deficit” or “threat to psychological safety” should be interpreted as interpreted themes derived from digital discourse rather than direct measurements of psychological states. These findings serve as hypothesis-generating insights into the public’s perception of healthcare resilience. We interpret this as a direct consequence of the 2024 healthcare crisis in South Korea, characterized by the mass resignation of resident doctors and the subsequent reduction in ED capacity [17]. Consistent with previous findings that prolonged waiting times significantly degrade patient satisfaction [18], our results confirm that the structural inability to access care, rather than the care itself, is the primary driver of consumer distress.

4.2 Pain and overtreatment: The dual burden of physical and psychological distress

The keyword analysis revealed “Pain” as the most frequent negative keyword, with “Overtreatment” ranking second. This indicates a dual burden on patients: the physical suffering from their condition and the psychological distress caused by distrust in the system. The prominence of “pain” aligns with research identifying pain management as a critical determinant of ED experience [19]. However, in the context of the current medical void, “pain” likely reflects the fear of untreated suffering due to delayed admission, rather than just the symptom itself. Furthermore, the emergence of “overtreatment” and “unnecessary” as top associated words highlights a deep-seated trust deficit. As noted in the literature, defensive medicine or uncertainty in diagnosis often leads to perceptions of overtreatment [20]. In the Korean context, this suggests that patients suspect hospitals of compensating for revenue losses during the crisis through unnecessary procedures, further eroding the patient-provider relationship.

4.3 Accessibility gaps: The “seoul-centric” and “holiday” dilemma

The analysis of time and place keywords revealed that ED searches spike during “Holidays” and “Nighttime,” and are heavily concentrated in “Seoul.” This underscores the fragility of the emergency care network during vulnerable times. The concentration of searches in Seoul, despite it having the highest density of medical resources [21], paradoxically indicates that even the best-resourced areas are struggling to meet demand during the crisis. This aligns with prior studies suggesting that the current “emergency room runaround” phenomenon is exacerbating regional disparities and creating bottlenecks even in metropolitan centers [22].

4.4 Implications for policy and practice

The findings offer critical implications for health information management and policy:

1. Crisis Communication Systems: The strong association between “medical void” and negative sentiment necessitates the implementation of real-time, transparent information systems that inform patients of ED capacity and wait times before they travel, reducing the risk of “runarounds.”
2. Restoring Trust: To address concerns of “overtreatment,” hospitals must prioritize transparent communication regarding diagnostic protocols and costs.
3. Resource Allocation: Policy interventions should focus on reinforcing ED staffing during high-demand periods (holidays, nights) and decentralizing emergency resources to alleviate the bottleneck in Seoul.

4.5 Limitations

This study has limitations. First, demographic variables (age, gender) could not be fully analyzed due to the nature of the aggregated text data. Second, the study did not distinguish between the types of EDs (e.g., regional emergency centers vs. local emergency rooms). Future research should employ panel data to capture individual-level variations in sentiment. Furthermore, as this study aggregated data from diverse platforms including news and SNS, the results may be influenced by ‘media amplification bias.’ Given that news articles constituted 40% of the total volume, journalistic tendencies to focus on systemic failures during national crises may have intensified the overall negative sentiment scores. Future research should integrate qualitative interviews with big data analytics to further validate these interpreted themes.