Infodemic insights: Mapping COVID-19’s digital discourse in Romania

The research data were systematically extracted using CrowdTangle, a platform that enables the collection of key engagement indicators for public Facebook posts CrowdTangle Team (2023). To ensure comprehensive coverage of the infodemic in Romania, we analyzed Facebook posts spanning from March 2020 to June 2022. This timeframe was selected because it not only captures the early outbreak phase but also extends to the point where detailed COVID-19 case reporting declined and public discourse around the topic began to subside.

To identify relevant posts, we employed the Romanian-language keyword “Covid” within CrowdTangle’s search function. This approach ensured that all selected content pertained to discussions surrounding the pandemic. The extracted posts include content from a diverse range of sources, including news outlets, government agencies, public figures, and individual users (Table 1).

The dataset used in this study consists of approximately 950,000 entries, each with 47 variables. These variables include the author’s name, post content, and metadata related to the user or page initiating the post. Additionally, the dataset captures various forms of engagement, such as the number and type of likes, shares, and comments, providing insights into interaction patterns.

The dataset focuses exclusively on Romania, with Facebook as the primary social media platform. This selection is based on Facebook’s dominant role in news dissemination and public communication, particularly by government authorities. As of 2024, Facebook maintained a 91% usage rate among Romanian social media users, with a potential advertising reach of 9.05 million users (Statista, 2024). Furthermore, Facebook dominates public news-sharing practices in Romania, serving as a significant platform for information distribution (Media Ownership Monitor Romania, 2024).

The second phase of this study involved sentiment analysis, which aimed to capture the emotional tone conveyed in Romanian Facebook posts related to COVID-19. By examining how different emotions were expressed in online discussions, this analysis helped identify fluctuations in public sentiment throughout the pandemic. To ensure accurate sentiment classification, we first applied several preprocessing steps to clean and standardize the text data. Since the posts were written in Romanian, we began by removing diacritics using the Unicodedata library in Python, converting special Romanian characters such as “ș” and “ț” into standard Latin script to ensure consistency.

Next, we performed tokenization using the Natural Language Toolkit (NLTK), breaking down posts into individual words to facilitate text analysis. All text was then converted to lowercase to maintain uniformity, and a custom vocabulary was created, ensuring the retention of pandemic-related terms while filtering out words that were unrelated to COVID-19.

To further refine the dataset, we applied stopword removal using NLTK’s Romanian language stopword list, eliminating common words such as conjunctions and pronouns that do not contribute meaningful sentiment. We also removed URLs, email addresses, and special characters (e.g., exclamation marks, question marks), as they lack semantic value. Additionally, emojis and hashtags were either removed or mapped to their text-based equivalents (e.g., → “happy”) to preserve their sentiment meaning. Further manual cleaning was conducted to eliminate incomplete words or anomalies that automated processes may have missed.

Once the text was preprocessed, sentiment analysis was conducted using the SpaCy library, which includes pre-trained Romanian language models capable of assigning sentiment scores to words based on their emotional value. Each term received a score ranging from −1 to 1, where −1 represents strong negativity, 1 represents positivity, and 0 indicates neutrality. The final sentiment score for each post was calculated by aggregating individual word scores, enabling classification into three primary sentiment categories: negative (−1 to −0.1), neutral (−0.1 to 0.1), and positive (0.1 to 1). Posts classified as negative often expressed frustration, fear, anger, or pessimism, while neutral posts presented factual content without emotional undertones. Positive posts, in contrast, conveyed optimism, relief, or encouragement.

Despite the rigorous methodology, one limitation of this approach is that some contextual nuances in Romanian language sentiment may not be fully captured by automated models, particularly in cases of sarcasm or implicit emotional expression. Additionally, because sentiment analysis relies on individual word scoring, longer posts containing mixed sentiment may not always be accurately classified, as positive and negative expressions may cancel each other out in the final sentiment score.

In the context of analyzing the COVID-19 infodemic in Romania, social media platforms, especially Facebook, provide a rich source of unstructured data. These data, when properly analyzed, can offer valuable insights into public sentiment, misinformation spread, and the role of influencers in shaping narratives. Given the volume of content generated during the pandemic, manual analysis is impractical. Researchers have thus turned to text mining techniques like Latent Dirichlet Allocation (LDA) to extract patterns and identify key themes related to COVID-19 (Thakur, 2022).

LDA, an unsupervised machine learning model, is particularly effective for analyzing unstructured text. It identifies hidden topics within large datasets by assuming that topics generate documents. LDA’s ability to recognize the multi-topic nature of social media posts makes it ideal for this study, allowing us to uncover themes related to public concerns, misinformation, and influencers’ roles in the COVID-19 discourse in Romania. The LDA model was used to analyze preprocessed Facebook posts, with five distinct topics emerging, including pandemic severity and vaccine discussions.

The final phase of this study applies SNA to examine how COVID-19-related information was disseminated on Facebook in Romania and to identify key influencers shaping public discourse. Given Facebook’s role as a primary platform for both public discussions and official announcements, understanding how information flows, who the most influential actors are, and how different communities interact is essential. To conduct this analysis, we used NetworkX, a Python library designed for analyzing the structure and properties of complex networks.

In this study, each Facebook page or user that posted COVID-19-related content is represented as a node, while interactions such as comments, shares, and mentions form the edges connecting them. The goal is to identify the most central nodes – those with the greatest influence on the network – since these actors play a critical role in shaping public opinion and determining how information spreads during the pandemic.

To quantify influence and network position, we calculated several key centrality metrics, which provide insight into the relative importance of different nodes in the network.

Degree centrality measures how many direct connections a node has to other nodes in the network, indicating its immediate influence based on its number of connections. It is calculated using the following formula: (1) C D ( v ) = deg ( v ) n − 1 , \begin{array}{c}{C}_{\text{D}}(v)=\frac{\deg (v)}{n-1},\end{array} where C D ( v ) {C}_{\text{D}(v)} is the degree centrality of node v, deg ( v ) \deg (v) represents the number of edges (connections) connected to node v, and n is the total number of nodes in the network.

A high degree centrality suggests that a node has numerous direct connections, making it a key player in disseminating information quickly across the network. In this study, Facebook pages or users with the highest degree centrality were identified as the most active hubs, distributing information broadly within the network.

Betweenness centrality measures how often a node acts as a bridge along the shortest path between two other nodes. This metric is crucial for identifying nodes that control the flow of information, as they serve as intermediaries between different clusters or groups within the network. It is calculated as: (2) C B ( v ) = ∑ s ≠ v ≠ t σ s t ( v ) σ s t , \begin{array}{c}{C}_{\text{B}}(v)=\sum _{s\ne v\ne t}\frac{{\sigma }_{st}(v)}{{\sigma }_{st}},\end{array} where σ s t {\sigma }_{st} is the total number of shortest paths from node s to node t, and σ s t {\sigma }_{st} is the number of those paths that pass through node v.

Nodes with high betweenness centrality are often gatekeepers or brokers of information, connecting different groups that may not be directly connected. These nodes are critical in the overall network structure, as they can influence the reach and spread of information by connecting otherwise disconnected parts of the network. In the COVID-19 discourse, these individuals or pages may have acted as conduits for information between various public and private organizations, news outlets, and general users.

Eigenvector centrality goes beyond simple connection counts by measuring not only the number of connections a node has but also the importance of the nodes to which it is connected. A node is considered influential if it is connected to other influential nodes. This is computed using the following formula: (3) x i = 1 λ ∑ j = 1 N A i j x j , \begin{array}{c}{x}_{i}=\frac{1}{\lambda }\mathop{\sum }\limits_{j=1}^{N}{A}_{ij}{x}_{j},\end{array} where x i {x}_{i} is the eigenvector centrality of node i, A i j {A}_{ij} is the adjacency matrix representing the connection between nodes i and j, and λ \lambda is the largest eigenvalue of the matrix.

Eigenvector centrality identifies nodes that, despite not having the highest number of direct connections, wield significant influence due to their links with other highly connected and influential nodes. In the context of the COVID-19 infodemic, this metric highlights pages or individuals who, while not the most frequent posters, are closely linked to key influencers and therefore have a broader reach within the network.

Beyond pinpointing influential actors, community detection algorithms were applied to uncover distinct clusters within the network. These clusters represent groups that interact more frequently with each other than with external sources, offering insight into how information was disseminated within specific communities. Mapping these interactions helps reveal which groups were most engaged with COVID-19 content and how different segments of the population processed and shared information. Certain political or medical figures may have held greater sway within specific communities, influencing the way information spreads and shaping public perceptions accordingly.

By examining degree, betweenness, and eigenvector centrality alongside community detection, we can better understand how COVID-19 information spread on Facebook in Romania. Identifying the most influential nodes in the network helps reveal who played a key role in amplifying or countering misinformation.

Before applying advanced analytical techniques, an initial EDA was conducted to uncover patterns, trends, and anomalies within the dataset. This process provides valuable insights and informs subsequent hypotheses for deeper investigation.

Figure 1 categorizes Facebook pages from Romania that posted about COVID-19 between 2020 and 2022 based on their level of activity.

Figure 1

Pages grouped by their level of activity in the top 30 categories.

The data reveal that political parties and politicians were among the most active contributors, emphasizing the centrality of COVID-19 in Romanian political discourse during the pandemic. This aligns with the broader trend of pandemic-related topics being leveraged for political narratives. Media organizations, including news sites, TV, and radio, also played a significant role, highlighting their function as key disseminators of COVID-19-related information. Additionally, community and non-profit organizations demonstrated considerable engagement, suggesting an effort to relay public health updates and resources to their respective communities.

Beyond these dominant categories, a variety of other page types – including educational institutions, entertainment pages, and personal blogs – actively participated in COVID-19 discussions, reflecting the pandemic’s far-reaching societal impact. The involvement of public figures and influencers further underscores the widespread diffusion of COVID-19-related discourse across diverse public and private spheres.

Figure 2 expands on this analysis by illustrating the number of views received by posts across different Facebook page categories within the same timeframe.

Figure 2

The number of views received by posts from different categories of Facebook pages related to COVID-19 in Romania.

The distribution of post views reveals that news sites were the primary source of COVID-19-related information for the Romanian public, followed by media news companies and government organizations. This indicates a strong reliance on official and journalistic sources for pandemic updates, consistent with expectations regarding public health communication.

Notably, government organizations garnered significant engagement, suggesting that official communications played a crucial role in public awareness and response strategies. The presence of categories such as newspapers and broadcasting media further underscores the pivotal role of traditional media in shaping public discourse. Conversely, categories such as educational sites, community organizations, and travel companies had comparatively lower viewership, suggesting that these sectors contributed less to the information ecosystem surrounding COVID-19 on Facebook.

Figure 3 provides insights into the Romanian public’s emotional engagement with COVID-19-related Facebook posts over time. The fluctuating patterns of reactions suggest distinct phases of sentiment throughout the pandemic.

Figure 3

Public’s reaction to Facebook posts over time.

The early peak in “Haha” reactions indicates that humor played a role in coping with initial uncertainty. Later, periodic spikes suggest that satire and irony remained a common response to policy changes and public discourse. The “Love” reaction peaking in mid-2021 aligns with potential moments of collective optimism, such as vaccine rollouts or community-driven support initiatives.

In contrast, the “Sad” reaction correlates with distressing events, such as high mortality rates or economic hardships, while “Angry” reactions rise during periods of public discontent, possibly reflecting backlash against government policies, lockdown measures, or vaccine mandates (Radu, 2021). The co-occurrence of sadness and anger suggests that frustration often accompanied moments of grief, reinforcing public dissatisfaction with crisis management.

Ultimately, the figure highlights the emotional volatility of the public’s response to the pandemic.

After processing the text and isolating tokens, we generated a WordCloud (Figure 4) to visualize the 100 most frequently used words in Facebook posts related to COVID-19. The predominance of terms such as “Covid,” “people,” “cases,” and “Romania” suggests that the discussion was primarily centered on the pandemic’s impact at both the national and regional levels. The presence of county names, references to vaccines, and medical terminology indicates a strong focus on public health concerns and the incidence of cases across different areas. Many of these geographical terms appear in conjunction with COVID-19, reinforcing the importance of location-specific discussions, as noted in previous studies (Papapicco, 2020).

Figure 4

WorldCloud of top 100 most used words extracted from Facebook posts.

The prominence of terms such as “persoane” (people), “cazuri” (cases), “coronavirus,” and “urgentă” (emergency) suggests that discussions were primarily centered on infection rates, mortality, and the severity of the crisis. The frequent appearance of words like “deces” (deaths) and “spital” (hospital) indicates that hospital capacity and fatalities were major public concerns, reflecting anxieties about Romania’s healthcare system.

The presence of geographical terms such as “București,” “Cluj,” and “Constanța” highlights that pandemic discussions were not only national, but also regional, likely in response to local outbreaks and policy measures. This reinforces the idea that the public closely followed how different areas were affected by COVID-19 and how authorities responded at a regional level. Additionally, the recurring mention of “vaccinare” (vaccination) and “împotriva” (against) points to significant engagement with vaccine-related discussions, reflecting both advocacy for immunization and the presence of vaccine hesitancy or opposition.

Another key observation is the focus on governmental measures, as indicated by the frequent use of words such as “măsuri” (measures) and “protecție” (protection). This suggests that public discourse was heavily influenced by discussions about pandemic-related policies, restrictions, and health guidelines, which aligns with global trends where government interventions became a focal point of debate.

To analyze sentiment trends, we used the SpaCy library, which applies a predefined Romanian-language dictionary to classify words as positive, negative, or neutral. Sentiment scores range from −1 (strongly negative) to 1 (strongly positive), while 0 represents neutrality. This classification allows for an assessment of the emotional tone within Facebook posts discussing COVID-19. Figure 5 provides a comparative analysis of newly reported COVID-19 infections and the public sentiment expressed in social media discourse over time.

Figure 5

COVID-19 cases vs (a) Polarity and (b) Sentiment extracted from Facebook posts.

In Figure 5a, the purple line represents the number of new COVID-19 cases, while the green line illustrates the polarity of Facebook posts, which measures the strength and nature of sentiment, ranging from negative to positive. The observed fluctuations in polarity align with key moments in the pandemic, potentially reflecting collective emotional responses to major events.

Figure 5b compares the new COVID-19 cases (blue line) with the sentiment extracted from Facebook posts (red line) over the same period. A sharp decline in sentiment at the beginning of 2020 is consistent with studies that indicate an initial wave of optimism or resilience at the onset of the pandemic, followed by increasing negativity as the crisis deepened (Fernandez et al., 2022).

Notably, there appears to be an inverse relationship between the sentiment extracted from social media and the number of COVID-19 cases. In periods where case numbers surged, public sentiment tended to become more negative, likely due to concerns over rising infections, hospitalizations, and restrictive measures. Conversely, sentiment improved during phases where case numbers declined, reflecting moments of relief or optimism.

These findings suggest that public sentiment on social media closely mirrored the real-world progression of the pandemic, emphasizing the role of online discourse as a real-time reflection of collective mood and crisis perception.

Table 2 shows the distinct topics with associated words and their respective probabilities within that topic.⁽¹⁾

The analysis of COVID-19-related Facebook discussions in Romania using LDA reveals distinct thematic clusters that structure the pandemic discourse.

The first topic centers on COVID-19 case reporting, emphasizing the number of cases, affected individuals, and fatalities. Key terms such as “covid,” “cazuri” (cases), “persoane” (people), “decese” (deaths), “pacienti” (patients), and “judetul” (the county) highlight a strong focus on official statistics and media updates tracking the pandemic’s progression. These findings align with broader research on crisis communication, where case metrics play a central role in shaping public perception (Papapicco, 2020). However, the dataset does not explicitly reflect discussions on trust in institutions providing these statistics, despite research suggesting that perceptions of data transparency influence compliance with health measures (Asif et al., 2021).

The second topic addresses the financial implications of the pandemic, including costs associated with testing, medical services, and quarantine accommodations. The presence of terms like “lei” (Romanian currency), “euro,” “tariful” (the rate), and “hotel” suggests that economic concerns were a significant part of pandemic-related discussions. This is consistent with global trends where financial anxieties became a major driver of online engagement, as highlighted in prior studies on COVID-19’s economic impact (Asif et al., 2021). The findings indicate that economic discourse may have been more emotionally charged than case reporting, particularly during lockdown periods and economic downturns. Additionally, the framing of financial burdens could have implications for vaccine hesitancy, as individuals weighed the perceived costs and benefits of immunization.

The third topic focuses on Romania’s public health response, including vaccination and containment measures. Key terms such as “vaccin” (vaccine), “impotriva” (against), “sanatatii” (of health), and “trebuie” (must) suggest a discourse centered on vaccine administration and compliance with health directives. Unlike studies that emphasize strong vaccine skepticism in Eastern Europe (Golos et al., 2023), the dataset does not prominently feature misinformation-related terms, indicating that the discussions were framed around practical implementation rather than ideological resistance. However, the lack of a dedicated family well-being topic, as seen in other pandemic-related analyses, suggests that these posts primarily focused on tracking pandemic developments rather than personal experiences. This distinction may stem from differences in data characteristics; while our dataset comprises posts focused on pandemic progress, other studies have examined how individuals sought to mitigate the pandemic’s effects on daily life.

The diversity of these thematic clusters highlights the multifaceted nature of online pandemic discourse, where public attention was divided between tracking infection data, navigating economic challenges, and discussing health interventions. However, merely identifying topics does not fully explain why certain themes dominated or how they evolved over time. The patterns observed in these discussions suggest that narrative structures played a key role in shaping public perceptions, reinforcing the idea that economic survival, government control, and scientific trust were central narratives influencing engagement with pandemic-related content (Shiller, 2017).

While the LDA model successfully delineates thematic boundaries, its results should be interpreted in the context of narrative economics, which argues that public behavior is influenced not only by raw data but also by the stories people construct around those data. The absence of explicit engagement with trust and misinformation narratives, despite their known impact on crisis responses, suggests that further research should investigate how sentiment, institutional credibility, and political discourse interact with these thematic categories. Additionally, expanding the comparative scope to include other Eastern European countries could provide insights into how Romania’s COVID-19 discourse differed from or aligned with broader regional trends, particularly concerning vaccine acceptance, economic anxieties, and political trust.

These findings also highlight the need for more targeted public health messaging strategies, as different aspects of the pandemic resonated with different segments of the population. Government agencies and health authorities could benefit from real-time monitoring of online discourse to identify emerging narratives and adjust their communication strategies accordingly. In particular, misinformation monitoring, artificial intelligence (AI)-driven content moderation, and community-based fact-checking could help counteract misleading narratives while reinforcing trust in official sources. Addressing economic concerns through transparent policy communication may also help mitigate resistance to health interventions, especially among those most affected by financial instability.

SNA is a methodological approach that applies statistical and mathematical tools to examine patterns of connections within social structures. It is used to identify key actors, understand communication dynamics, and analyze the overall structure of networks consisting of individuals or entities. A core focus of SNA is the study of network characteristics such as centrality, connectivity, and community structures, as well as the diffusion of information or behaviors through the network. This method helps determine how information propagates and through whom it spreads, offering valuable insights into the role of influential nodes in the network. Visualizing these structures through graphs further clarifies relationships, highlights communication patterns, and identifies weak ties relevant to research and intervention strategies.

To conduct this analysis, we employed Python’s NetworkX library, which facilitates the modeling of complex social networks and the identification of critical relationships and influencers. This is particularly relevant in the context of a pandemic, where understanding the spread of information, misinformation, and public discourse is crucial. SNA considers both the fundamental components of a network – nodes (representing entities such as Facebook pages or users) and edges (representing interactions between them, such as mentions, comments, or shares) – and the nature of these relationships, whether directed, undirected, or weighted. Applying this method requires several steps, including data processing and natural language processing (NLP) techniques to extract meaningful insights from large volumes of social media data.

4.4.3

Network graph construction

Once the dataset was refined, the NetworkX library was used to transform these relationships into a structured network graph. The weights assigned to the edges were incorporated as attributes, visually representing the strength of connections between nodes. This approach enabled the identification of the most influential actors in the Romanian COVID-19 discourse on Facebook, offering insights into information flow, community formation, and potential misinformation dynamics.

Figure 6 presents a graphical representation of the network, where nodes represent individuals or entities, and the edges between them indicate relationships or interactions. The clustering of nodes reflects communities or subgroups, while highly connected nodes serve as key influencers or central hubs of activity.

Figure 6

Social Network graph.

To analyze the network, we examine various metrics that offer insights into its structure and dynamics, helping to understand how information flows, which nodes play a critical role, and how resilient the network is to disruptions. One such measure is degree centrality, which identifies the most connected nodes. These highly connected individuals or entities are likely to be influential in spreading information, well-informed, or capable of rapidly reaching a broader audience within the network.

Figure 7 provides insights into the most influential entities in the COVID-19 discourse on Facebook, ranking the top 30 nodes by their degree of connectivity. The data reveal a hierarchical structure of information dissemination, where political figures and media entities serve as central actors in shaping public narratives about the pandemic.

Figure 7

Degree of centrality, top 30 nodes in the network.

Klaus Iohannis, the President of Romania, emerges as the most frequently mentioned individual, indicating his centrality in discussions surrounding pandemic response and policy decisions. He is followed by Ludovic Orban, Romania’s Prime Minister from November 2019 to December 2020, and Vlad Voiculescu, the Minister of Health during the pandemic. Their similar levels of centrality suggest a comparable degree of influence in shaping public discourse, particularly as figures associated with government crisis management efforts.

A key institutional figure, Raed Arafat, the Secretary of State and Head of the Department for Emergency Situations, holds a prominent position within the network. As the primary government communicator on restriction measures and emergency protocols, Arafat’s high connectivity reflects the public’s reliance on authoritative sources for pandemic updates. The visibility of emergency response officials aligns with studies on crisis communication, where official health authorities play a pivotal role in structuring online discussions and public sentiment (Yang et al., 2021).

Among media entities, Digi 24 emerges as the most central television network, reflecting the dominant role of mainstream news outlets in pandemic-related discourse. This finding supports previous research showing that traditional media, despite competition from social media platforms, remained a key conduit for COVID-19 information dissemination (Papapicco, 2020). Cronica.ro follows as another highly cited news source, suggesting a strong engagement with digital media platforms. The presence of USR Plus as the most central political party in discussions highlights its salience in public debates, likely due to its policy stance during the pandemic and active involvement in governance during the 2019–2020 period.

The results underscore the concept of “superspreaders,” where a small number of highly connected nodes exert disproportionate influence over information flows (Yang et al., 2021). In the Romanian context, political leaders and major news organizations function as key amplifiers, shaping the dominant narratives that structured public understanding of the pandemic. However, these findings also raise important questions about the dynamics of information control, the potential for misinformation diffusion through authoritative channels, and the role of social media in reinforcing or challenging official narratives.

The analysis of network centrality measures highlights the key actors in Romania’s COVID-19 discourse, revealing how information circulated and which entities played the most significant roles in shaping public narratives. By examining Betweenness Centrality, Closeness Centrality, and Eigenvector Centrality, we can distinguish between those who acted as bridges, rapid disseminators, and structurally influential figures within the network.

Betweenness Centrality, as shown in Figure 8, identifies nodes that serve as intermediaries between different parts of the network, meaning they control the flow of information between otherwise disconnected groups. Klaus Iohannis, Romania’s President, stands out as the most central figure, reinforcing his role as a key point of reference in pandemic-related discussions. His high betweenness centrality suggests that information on COVID-19 was frequently channeled through him, amplifying his role as a dominant voice in the digital conversation. Alexandru Rafila, Romania’s representative in the WHO Steering Committee until 2021, and Cronica.ro, a major digital news source, also rank highly, indicating their strategic position in information diffusion. Compared to Marcel Ciolacu,⁽²⁾ these entities appear to have played a more pivotal role in connecting different segments of the discourse.

Figure 8

Betweenness centrality, top 30 nodes.

Closeness Centrality, presented in Figure 9, measures how quickly a node can reach all others in the network. The top-ranking entities remain relatively stable, reflecting the high density of interactions within the dataset. However, one notable shift is that Gabriela Firea (former mayor of Bucharest) replaces USR PLUS in the top 10, suggesting that while she may not be the most frequently mentioned figure, she is well-positioned to spread information efficiently across the network. This distinction highlights the importance of both direct visibility and structural positioning in information flow.

Figure 9

Closeness centrality, top 30 nodes.

Eigenvector Centrality, illustrated in Figure 10, measures a node’s influence based on the centrality of its connections, meaning that being linked to other highly influential figures increases a node’s overall importance. Klaus Iohannis leads once again (0.2666), followed closely by Ludovic Orban (0.2468) and Vlad Voiculescu (0.2447). Their consistently high rankings across different centrality measures confirm their dominant role in structuring pandemic discourse. Digi24 and Raed Arafat, scoring 0.2249 and 0.2248 respectively, reinforce the strong presence of media organizations and emergency management authorities in shaping public communication. Gabriela Firea and Antena 3 rank lower but remain prominent, reflecting their sustained relevance in COVID-19 discussions.

Figure 10

Top 30 eigenvector centrality scores.

These findings illustrate the hierarchical nature of information dissemination in the Romanian social media landscape, where political leaders, health officials, and major news outlets played crucial roles in shaping narratives and public perception. The persistence of certain figures across all three centrality metrics suggests that they were not only highly visible but also structurally embedded within the network, influencing the trajectory of information exchange.

While centrality measures help map the architecture of pandemic discussions, they also raise important questions about the implications of centralized information flows. The dominance of political figures and institutional actors suggests that the framing of COVID-19 policies and public health measures was largely controlled by a small set of influential entities.