Dynamics of Networked Framing: Automated Frame Analysis of Government Media and the Public on Weibo With Pandemic Big Data

Abstract

Networked framing focuses on how the public becomes gatekeepers on social media. To unpack the dynamics of networked framing, we conducted an automated frame analysis to identify the shift of frame structures of government media (N = 12,090) and the public (N = 1.49 million) on Weibo during the COVID-19 pandemic. We found a moderate level of frame alignment between government media and the public, with high divergence observed during the pandemic’s initial stage. The public challenged government media frames by deploying unique frame functions and creating new frames, but their frame network was fragmented relative to that of government media, which constructed a cohesive network of frames to enhance discursive control.

The proliferation of information and communication technologies (ICTs) and the increasingly interconnected public have challenged the pervasive role of legacy media and diversified voices in public communication globally (Bennett & Pfetsch, 2018). Communication scholars have increasingly examined how organizations and individuals negotiate the frames of impactful events on social media (Nip & Fu, 2016b; van der Meer et al., 2014). Networked framing is the process in which frames are constantly revised, rearticulated, and redistributed by elites and the crowd on social media (Meraz & Papacharissi, 2013). Initially, the literature focused on the crowd’s growing role as gatekeepers and how Twitter’s affordances (e.g., hashtags) allowed non-elites’ narratives to gain prominence. Subsequent studies have identified factors that condition networked framing in public and counterpublic spheres in various contexts, such as polarization (Pöyhtäri et al., 2019), censorship (Jiang et al., 2015), or construal level (Kwon et al., 2017).

This study contributes to the thriving literature in two important ways. First, traditional framing research focused on how audiences reacted to a discrete frame advocated by an elite actor (Borah, 2011). Recent scholars have questioned the external validity of such experimental framing research, as individuals typically consume multiple or even competitive frames simultaneously in reality (Nisbet et al., 2013; Zhao & Oh, 2021). Previous studies have shown that the interrelationships between different messages can amplify or undermine the effectiveness of a certain frame (Anthony et al., 2013; Nisbet et al., 2013). Thus, the literature can benefit from a network approach emphasizing a constellation of frames and the mutual influence between different actors’ frame networks.

Second, despite the “fluidity and transience” of the iterative process of networked framing (Meraz & Papacharissi, 2016, p. 99), there has been relatively limited attention on the dynamics of networked framing enacted by elite and non-elite actors (cf. Knüpfer et al., 2020). To capture the complex dynamics of different actors’ framing in a turbulent media ecology, we use frame alignment (i.e., the extent to which different actors demonstrate similarity in framing) and frame diversity (i.e., the heterogeneity of frames for each actor). Whereas frame alignment allows media and the public to negotiate meanings and coordinate collective responses (Gerken & van der Meer, 2019; van der Meer et al., 2014), public frame diversity, observable in people’s creation and amplification of frames counter to elite ones, allows the public to demonstrate discursive resistance and contest what issues mean (Kwon et al., 2017; Molina, 2019). Measuring the shift of frame alignment and frame diversity helps unpack the dialectical relationship between government media and the public in various social media contexts.

We collected big data from Weibo during the COVID-19 pandemic. Weibo, among the top social networking sites in China, is an indispensable part of the everyday lives of Chinese people. In particular, we performed an automated analysis of frame structures, interrelationships, and dynamics of government media and the public during the pandemic. In an authoritarian society like China, government media, a kind of elite media, remarkably shape public opinion. This study enables us to form a more coherent understanding regarding the dynamics and fluidity of discursive relationships between an authoritarian government and ordinary citizens. Our study also enriches the literature on networked framing by taking an ecological, systemic, and dynamic perspective on such a complex process.

Literature Review

Framing

Literature on framing increasingly offers a powerful perspective for understanding news frames and their impact on how audiences perceive issues (de Vreese, 2005). To date, communications scholars have typically scrutinized two types of frames (Chong & Druckman, 2007): news frames, which represent how news media focus on specific aspects of issues and thus promote particular interpretations of them, and individual frames, which refer to individuals’ perceptions of given issues.

Gamson and Modigliani (1989) have defined news frames as interpretive packages composed of reasoning devices such as exemplars, metaphors, and visual images. At the core of an interpretive package, a frame allows communicators to organize issue discourse by selecting and excluding elements to make meaning of the issues. As such, frames “promote a particular problem definition, causal interpretation, moral evaluation, and/or treatment recommendation” for an issue (Entman, 1993, p. 52). Benford & Snow, 2000 also discussed three core functions of frames: diagnostic (i.e., problem definition), prognostic (i.e., proposed solution), and motivational (i.e., call to action).

Considering country-based differences in media control and freedom of expression, some scholars have additionally proposed unique frame functions in authoritarian regimes such as China. They contended that value advocacy (e.g., the promotion of nationalism or “positive energy”) has been an important frame function used by the authoritarian party-state to maintain hegemony and gain legitimacy (Gries et al., 2016). In the context of social media, scholars have also emphasized the frame function of emotional expression among the public (Papacharissi, 2016).

During public health crises, framing allows organizations such as public health agencies, associations, and corporations to define situations and set the stage for public debate (Liu & Kim, 2011; Zhao & Oh, 2021). The literature focuses on how elite media such as legacy media or government media use discrete frames during crises (J. Li et al., 2017; Ngai et al., 2020; Shih et al., 2008). For example, Shih et al. (2008), using The New York Times as a case, found that consequence (e.g., cases of infection) and crisis response action were the most prominent frames during three epidemics (e.g., avian flu). Ngai et al. (2020) found that in covering COVID-19 in China, government media like People’s Daily relied heavily on action and new evidence frames. Meanwhile, research on the dynamics of media frames during crises has been limited (cf. Shih et al., 2008). Thus, the literature would benefit from research on how elite media and the public co-construct multiple frames over time.

Networked Framing

By allowing affordable, interactive, instantaneous, and many-to-many communication, social media enable people to develop distinct interpretations of issues and share their individual narratives with their followers in the networked public sphere. As important sources of information, public discussions on social media can affect people’s understanding and consumption of information about crises. Empirical studies have shown that public framing on social media plays a particularly significant role during critical events (Kwon et al., 2017; Masip et al., 2019; van der Meer, 2018).

Networked framing recontextualizes framing in the complex information environment characterized by interactions among mainstream media, alternative media, and online/social media (Bennett et al., 2018). As a crowd-centered theory, networked framing represents a shift from “top-down, prior control paradigms to crowd-centered, networked, and collective intelligence logics.” (Meraz & Papacharissi, 2016, p. 3) As social media like Twitter or TikTok connect dispersed users through collaborative gatekeeping, both elite (e.g., mainstream media, organizations, politicians, influencers) and non-elite (e.g., unofficial sources, ordinary people) actors impact how social issues are interpreted by creating content, sharing certain news items, and engaging in horizontal conversations (Bennett et al., 2018).

Networked framing is characterized by three interrelated processes (Meraz & Papacharissi, 2016): (a) collaborative filtering, enabled by platform algorithms that customize filtered content based on personal connections, content relevance, and how other users engage with the message; (b) power law of participation that allow a few social media users to become prominent by disproportionately winning the attention and support of the crowd through social acts, such as sharing, liking, and hashtagging, in a competitive process; and (c) the homophily principle, which drives individuals to tend to share the content of similar views to amplify their preexisting perspectives. In these processes, social media’s adhesivity and conversational markers (e.g., retweets and hashtags) sustain the flow of information and afford a plurality of narratives by giving voice to marginalized publics and issues (Knüpfer et al., 2020).

Recent studies have revealed certain characteristics affecting networked framing, such as polarization (Pöyhtäri et al., 2019) or construal level (Kwon et al., 2017). Pöyhtäri et al. (2019) showed networked framing can lead to political polarization during the “refugee crisis,” as conservative and right-wing groups use social media to spread and amplify their anti-immigration views among their supporters. The authors commented that there is “nothing inherently democratizing or pluralizing about networked framing as a citizen activity” (p. 19). Given that the effect of networked framing can be highly contextual, we studied networked framing in a controlled media environment.

Networked framing in a controlled environment

There are three views regarding how networked framing shapes the discursive sphere in an authoritarian society such as China. First, social media can serve as public forums for alternative voices, in which governments exercise less control and censorship than in traditional media (DeLisle et al., 2016). The empowering role of networked framing should increase the diversity of public opinion, mobilize collective actions among counterpublics, and ultimately foster democracy in authoritarian regimes (Chan, 2018; Nip & Fu, 2016a). Indeed, recent studies have revealed the increasingly important role of non-elite actors in gatekeeping and framing on social media in non-democratic countries (Jiang et al., 2015).

Second, the authoritarian government can use social media to further control society and strengthen censorship. In authoritarian societies, social media operate in sophisticated, complex systems of digital control involving control over access, the censorship of content and/or accounts, and the fabrication of public opinion (King et al., 2013; Sullivan, 2012). For instance, content on Weibo has been actively censored to comply with government regulations (Sullivan, 2012). Studies have also shown that, in China, social media use is generally associated with nationalism and support for the existing political system (Gan et al., 2017; X. Li et al., 2016), suggesting resonance between government media frames and public frames on social media.

Third, the implication of networked framing is contextual and fluid, as the public and the authoritarian government engage in a tug-of-war to negotiate and contest the dominant discourse in a specific context. There has been growing evidence supporting the existence of fragmented, temporary public spheres in China concerning certain issues on social media, such as environmental protection or food safety (Rauchfleisch & Schäfer, 2015; Wang & Song, 2020). These findings contribute to our understanding of networked framing in a tightly controlled information environment and call for more research on the dynamic relationship between government media and public frames in an authoritarian society.

The Relationship Between Government Media and Public Frames on Social Media

As the shifting relationship between government media and public framing during the pandemic remains unclear, we sought to elucidate the dynamic relationship between government media frames and public frames on social media. The COVID-19 pandemic provides a suitable context for capturing the fluidity and transience of networked framing, as this ongoing crisis is extremely impactful, dynamic, and involves multiple events on which the government and the public can contest the dominant discourse on Weibo. We formulated two research questions on frames and frame functions of government media and the public:

• RQ1: What government media and public frames have emerged on Weibo during the pandemic?

• RQ2: What functions do government media and public frames exhibit on Weibo during the pandemic?

Frame alignment

According to Schultz and Raupp (2010), crises trigger “sensemaking and sensegiving processes” in which actors engage in framing to impact the crisis interpretations and reality constructions (p. 113). Frame alignment measures the extent to which different actors demonstrate similarity in framing over time (Gerken & van der Meer, 2019). Frame alignment thus captures the dynamic, evolutionary nature of phenomena as frames of actors interacting with each other result in mutual borrowing, convergence, and eventual alignment (van der Meer et al., 2014). In communication about crises, different actors tend to align frames to resolve conflicting understandings of crises, co-construct their crisis-related experiences, and coordinate critical measures in response (Zhao & Oh, 2021).

Previous research suggests a plausible association between government media and public frames on Weibo. On the one hand, media organizations are still influential and can directly impact public discourse, particularly in an authoritarian society (Siapera et al., 2018). Nip and Fu (2016a) argued that the importance of China’s news organizations as agenda/frame setters has been transferred online. On the other hand, social media can be a site of discursive contestation by disseminating counterdiscourses through “jamming” (Chan, 2018; Jackson & Welles, 2015). Chan (2018) found that early posts on the Hong Kong Police Force Facebook Page were heavily jammed with various kinds of counterdiscourses, but the intensity of the counterdiscourses decreased over time. Thus, there should be a moderate level of alignment between government media and public frames during the pandemic.

Moreover, frame alignment among different message sources is paramount to facilitating the public’s behavioral change (e.g., wearing facemasks) and eventually mitigating the crisis (e.g., stopping the viral spread). This is because the public often evaluates information credibility and determines preventive behaviors based on the convergence of messages from multiple sources (Anthony et al., 2013). Thus, as the pandemic evolves, some level of frame alignment between actors on social media should emerge for the resolution of the pandemic. In other words, the magnitude of frame alignment should increase over time. We propose the following hypothesis:

• H1a-b: Government media frames are associated with public frames on Weibo throughout the pandemic (a), and the magnitude of the association increases during the pandemic (b).

Beyond examining the alignment of discrete frames, our study supplements the literature by considering the alignment of frame networks over time. A social network approach to framing considers frames as clusters of frame elements in semantic networks (e.g., Leydesdorff & Hellsten, 2006). In such networks, nodes are frame elements linked by certain kinds of semantic relationships, such as the co-occurrence of those elements (Walter & Ophir, 2019). That approach offers a refined understanding of a constellation of frames and their relationships to a given issue and enables the dynamic comparison of the configuration and convergence of frames by different actors. Thus, we developed a research question regarding the change of frame networks during the pandemic:

• RQ3: How do frame networks of government media and the public change on Weibo throughout the pandemic?

Frame diversity

Diversity is the foundation for informed citizenry and pluralistic public discourse. Previous media studies have solely focused on source or content diversity (Loecherbach et al., 2020), and there has been insufficient research regarding frame diversity, the heterogeneity of frames in news and/or public discourse (Huang, 2010). According to Chong and Druckman (2007), “the role of multiple competing frames” (p. 101) has not been well understood. Considering the important implications of frame multiplicity, we call for more scholarly efforts to compare frame diversity among different actors and investigate the dynamics of frame diversity in the social media context.

Reframing occurs when people produce and disseminate frames that diverge from elite narratives (Chan, 2018; Jackson & Welles, 2015; Knüpfer et al., 2020). Reframing should generate higher frame diversity in the public sphere, as it allows the public to create alternative frames challenging elite narratives. For example, Nip and Fu (2016b) examined how Weibo users construct frames by reposting corruption-related content. They found that although some reposts of news organizations only repeated the frames in the source posts, other reposts contained new frames and frame functions discrepant from the official discourse on corruption. This suggests that Weibo’s reposting function enables users to construct and disseminate alternative frames, thereby increasing the diversity of public frames in the issue arena. Pöyhtäri et al. (2019) showed that there were more perspectives regarding the refugee crisis on social media than mainstream media. Thus, during the pandemic, the diversity of the public frames should be higher than that of government media frames on Weibo.

From a dynamic perspective, frame diversity should also fluctuate due to political, technological, and socioeconomic factors as crises unfold. That view represents a more systematic approach to understanding networked framing as both an outcome and process of framing and reframing practices by actors on social media. Due to the ebb of attention in the resolution stage of a pandemic, frame diversity of the public and government media should both decrease over time. Thus, we propose the following hypothesis:

• H2a-b: The diversity of the public’s frames is higher than that of government media frames (a), and their frame diversity decreases during the pandemic (b).

Scholars have called for additional research on how media diversity influences perceptions of diversity in the public (Huang, 2010; Zhao, 2016). To date, studies have shown that when media provide more diverse content about an issue, audiences have access to a greater variety of personally applicable frames, which results in more diverse public frames (Huang, 2010; Peter & De Vreese, 2003). By extension, the wider availability of alternative narratives can supplement issue interpretations and promote alternative solutions (Leach et al., 2010). On Weibo, diversity in government media can also create a heterogeneous opinion climate and reduce normative pressure to follow dominant opinions, thereby increasing people’s willingness to construct new frames. In sum, we propose the following hypothesis:

• H3: The diversity of government media frames is positively associated with the diversity of public frames.

Method

We conducted an automated frame analysis to investigate government media’s posts and the public’s reposts on Weibo during the COVID-19 pandemic. Weibo, with functions resembling Twitter’s, has approximately 566 million monthly active users (China Internet Watch, 2021). It is the most popular microblogging website in China and deeply affects the everyday lives of Chinese people. Thus, discussions about the pandemic on Weibo provide a unique opportunity to study the dynamic interactions between government media frames and public frames. Government media represent both the mouthpiece of the party-state and the voice of people in China and are thus the most influential elite media in China.

Specifically, we identified frames and frame functions using pandemic big data by integrating the inductive and deductive approaches. First, we extracted the topics presented in the media posts and public reposts through latent Dirichlet allocation (LDA) topic modeling. Those topics (i.e., frame elements) were inductively analyzed by the researchers to discover frames. Frame functions were then deductively coded based on the literature. Last, frame alignment and frame diversity were used to measure the dynamics of networked framing on social media.

Data Collection

The time frame is from January 1, 2020, when COVID-19 was first reported, to April 8, 2020, when new cases had significantly decreased in China. Relevant contents were identified by conducting a keyword search of “coronavirus” and its variant form in Chinese and subsequently downloaded using a Python web crawler, along with publicly visible information about the posts and their senders (see Supplementary Information [SI] P1-3 for details). Data were downloaded between February and April 2020. As a result, we collected a total of 13.79 million Weibo posts/reposts/comments (along with their metadata), of which 5.34 million were reposts (38.7%). These data constituted a pool from which the public’s reposts were extracted.

For government media posts, following the suggestions of professional journalists, we selected five highly representative government news media in China: People’s Daily, CCTV News, Global Times, Xinhua News Agency, and China News Service. After retrieving all relevant news (N = 12,090) authored by those media outlets, we extracted the reposts of those posts from the pool, with reposts from media and government accounts removed, and ended up retrieving approximately 1.49 million public reposts. Given the tremendous amount of public discussion on the topic, we did not crawl all reposts. To validate the crawled sample’s representativeness, we compared the number of reposts shown on Weibo with the number of reposts collected for our sample. The sample size, approximately 5% of the population, correlated highly with the population (r = .98), which indicates its robust representativeness on discourse about COVID-19 on Weibo.

Automated Frame Analysis

We automatically identified different actors’ frames via an integrated inductive and deductive approach (van der Meer, 2018; van der Meer et al., 2014). Research relying solely upon semantic network analysis or topic modeling for automated framing analysis has faced certain challenges. For instance, topic modeling is an unsupervised machine learning algorithm to identify topics—that is, sets of words co-occurring frequently in texts (Blei & Lafferty, 2009). Topics detected by topic modeling might not be conceptually equivalent to frames (Maier et al., 2018). For a solution, we treated topics as frame elements and manually labeled and grouped those elements under larger themes, that is, frames (van der Meer, 2018; Walter & Ophir, 2019).

Figure 1 presents an overview of our automated frame analysis (see SI P4). First, after standard preprocessing procedures, we segmented the cleaned documents into terms, which we subsequently transformed into a document–term matrix. Second, we employed LDA to automatically model topics in the documents. For media posts and public reposts, we conducted separate LDA analyses. Following standard procedures (Blei & Lafferty, 2009), we determined the number of topics to be 40 and set the alpha value at .1 for both LDA analyses (Figures S1–S2). Next, we conducted social network analysis and community detection. Namely, we calculated similarity matrices between topics using cosine similarity based on the topic–term distribution matrices, which we later transformed into networks with nodes representing topics and edges representing degrees of similarity. Last, we manually labeled topics (N = 80) into frames and frame functions using the results of community detection and topic modeling (Figure S3). For valid interpretations of frames, we conducted a textual analysis to label frames inductively (Table 1). Using theory-informed coding rules (Table 2), we also conducted content analysis to code frame functions deductively (for details, see SI P5).

Figure 1.

A visualization of the procedure of automated framing analysis.

Note. LDA = latent Dirichlet allocation.

Table 1.

Frames in Government Media Posts and Public Reposts.

Frame	Description	Media posts	Public reposts
International public health outbreak	Global cases of infection and/or mortality, disease spread, or outbreak severity	22.5%	3.2%
Domestic public health outbreak	Domestic cases of infection and/or mortality, disease spread, or outbreak severity	23.4%	3.0%
Public health response	Deployment of health resources, the health system, or health professionals’ work	8.6%	6.4%
Public health education	Education about disease transmission, prevention, or treatment	4.3%	8.6%
Government regulations	Regulations about the economy, food supply, transportation, or quarantine	32.4%	9.3%
Consequences	Consequences of the pandemic (e.g., cancelation of events or travel plans)	7.1%	—
Detection and testing	Methods or results of COVID-19 testing	4.3%	—
Research and treatment	Scientific research and discovery about disease treatment and vaccines	5.3%	3.4%
Foreign actors	Foreign actors’ commentary on the pandemic and the government’s role	2.0%	3.2%
Commemoration	Memories of people killed by the pandemic	1.2%	10.3%
Human interest	Personal stories or emotional angles on the issue	5.8%	45.3%
Information transparency	Transparency of information about donation, rumor labeling, or attribution of official responsibilities	—	5.7%
Call to action	Requests for public actions (e.g., preventive measures or dissemination of information)	—	12.3%
Other	Other frames with low percentages (e.g., Li Wenliang’s event, schadenfreude, and environmental protection)	—	2.6%

Note. The probability of each frame was recorded as 1 if ≥0.3 and as 0 if <0.3. Others (<1%) were excluded from subsequent analyses.

Table 2.

Frame Functions Presented in Government Media Posts and Public Reposts.

Frame function	Description	Media posts	Public reposts
Problem definition	Clarified key facts related to the problem, including regulatory/legal issue, institutional issue, public health issues, research issues, economic issues, etc.	61.2%	35.1%
Attribution	Moral judgment, blaming, conspiracy, or cover-up; used for assigning causal responsibilities or moral evaluations to certain actors who are responsible for the risks and consequences	11.4%	4.1%
Solution	Proposed solutions or actions	49.2%	34.8%
Value advocacy	Emphasized positive values, heroes, and state propaganda	27.1%	24.8%
Mobilization	Call to actions among the public and motivate the public to engage in action addressing self-quarantine, sterilization, and spread,	5.6%	4.7%
Emotional expression or support	Expressed emotions such as joy, hope, respect, anxiety, fear, schadenfreude or provided emotional support	8.9%	48.6%

Note. The probability of each frame function was recorded as 1 if ≥0.3 and as 0 if <0.3.

Derived Measures and Analytical Schemes

Frame diversity

To measure the frame diversity in media and public response, we employed the entropy measure (Shannon’s H, Shannon & Weaver, 1949), which has been widely used in communication studies to measure the multiplicity of states (Huang, 2010). Entropy measures the spread of observations across a defined number of categories, weighted by the occurrence of these categories.

Frame alignment

The interrelationship between media frame and public frame (i.e., frame alignment) was measured in three aspects: alignment of discrete frames, alignment of frame diversity, and alignment of frame networks.

Alignment of discrete frames

We measured frame alignment by comparing frames by government media and the public over time. Once media posts were matched with their public reposts, we established document–frame matrices for government media and the public by grouping topics in the document–topic matrices into frames. We measured the cosine similarity between the document–frame matrix of each news report and the document–frame matrix of its reposts, after which we averaged similarity scores to evaluate the alignment of discrete frames.

Alignment of frame diversity

To measure the alignment of frame diversity over time, we calculated the contemporaneous correlation between media frame diversity (i.e., the entropy of government media frames) and public frame diversity (i.e., the entropy of public frames).

Alignment of frame network

We determined the frame networks of government media and the public at three phases of the pandemic based on their paired occurrence within each (re)post. In each undirected, weighted co-occurrence network, we included a link between two frames (i.e., nodes) if both appeared in the same (re)post. We further divided the networks based on the three stages: initial (i.e., January 2020), middle (i.e., February 2020), and resolution (i.e., March and April 2020). After that, we analyzed the alignment of the government media and public frame networks at different stages via quadratic assignment procedure (QAP), a resampling-based method (similar to bootstrapping) for calculating network–network correlations. We also measured the centralization score of the networks, which describes the extent to which a network is organized around particular focal nodes.

Results

Government Media and Public Frames

In response to RQ1, which addressed government media and public frames on Weibo during the pandemic, our analyses revealed nine frames in news posts and public reposts (Table 1). Government media emphasized the frames of “Government regulation” (32.4%), as well as “Domestic public health outbreak” (23.4%) and “International public health outbreak” (22.5%), whereas public reposts tended to be framed with “Human interest” (45.3%), “Call to action” (12.3%), and “Commemoration” (10.3%). Spearman’s rank correlation coefficients revealed no significant relationship between frames in news posts and public reposts (r_s = −.25, p = .42). As such, frames emphasized by government media differed starkly from those of the public.

Frame Functions of the Public and Government Media

Table 2 summarizes results in answer to RQ2, which addressed the frame functions of government media and the public. Public reposts focused on the frame functions of emotional expression and support, whereas news posts emphasized problem definition and solution-related functions. At the initial stage, government media primarily used frames for problem definition, while the public primarily used frames for problem definition and solutions. As the pandemic developed, government media gradually began to focus on solutions, whereas in their reposts, the public focused on emotional expression (see Figure S4 for details).

Frame Networks of the Public and Government Media

RQ3 asked how the government media’s and the public’s frame networks changed on Weibo as the pandemic continued. Figure 2 illustrates the government media and public frame networks at the pandemic’s different stages. Initially, government media used the “Domestic public health outbreak” frame, juxtaposed with the “Government regulations” and “Detection and testing” frames. At later stages, government media not only continued to emphasize those frames but also began to employ the “International public health outbreak,” “Public health response,” and “Consequences” frames. At those stages, the “International public health outbreak” frame often coincided with other frames, including “Consequences,” “Government regulations,” and “Public health education.”

Figure 2.

Co-occurrence networks of media frames (top) and public frames (bottom) during three stages of the COVID-19 pandemic: (A) initial (left), (B) middle (middle), and (C) resolution (right).

Note. Node size is scaled to the weighted degree of the frames, while edge width is scaled to the weight of co-occurrence.

Regarding the public’s frame network, “Human interest” was persistently used across stages and often co-occurred with “Public health response” and “Commemoration” frames. In addition, “Public health education” and “Government regulations” were stressed at the initial stage, whereas “Call to action” and “Commemoration” were emphasized later on. Throughout the stages, “Information transparency” and “Call to action” were emphasized only by the public but not by government media.

The QAP revealed significant correlations among media frame networks at different stages (r_t1–t2 = .46, p < .01; r_t2–t3 = .68, p < .001; r_t1–t3 = .47, p < .05) and among public frame networks stage by stage (r_t1–t2 = .76, p < .01; r_t2–t3 = .63, p < .001; r_t1–t3 = .54, p < .05). However, at all three stages, none of the correlations between government media and public frame networks were significant. Furthermore, the degree of centralization was greater in media frame networks (C_D = .48) than in public ones (C_D = .41), which suggests greater frame cohesion (i.e., a lower level of fragmentation) for government media than the public.

Frame Alignment and Frame Diversity

H1a assumed that government media frames were associated with public frames on Weibo throughout the pandemic. In response, our analysis revealed moderate alignment between government media frames and public frames (M = .52, SD = .27). Figure 3A, which visualizes the alignment of discrete frames, shows the large variance of alignment at the initial stage. These results support H1a.

Figure 3.

Dynamics of frame alignment, media frame diversity, and public frame diversity: (A) alignment of frames between media and public, (B) diversity of media frame, and (C) diversity of public frame.

Note. Frame alignment between media and the public (a value ranges from 0 to 1) was measured as the cosine similarity between the document–frame matrix of each news post and the document–frame matrix of its reposts. Frame diversity was measured as the entropy (Shannon’s H) of media frames and public frames. Regression lines in graphs present the linear relationship between y-axis (alignment or diversity) and time.

H2a proposed that the diversity of the public frames was higher than that of government media frames. Figure 3B and C presents the dynamics of media and public frame diversity throughout the pandemic. Our Welch’s t-test revealed that public frames showed a higher level of diversity (M = 1.82, SD = 0.33) than media frames (M = 1.03, SD = 0.37), t(149.21) = 39.28, p < .001, leading to the support of H2a.

To test H3 regarding the association between government media frame diversity and public frame diversity, we analyzed the correlation between government media and public frame diversity. Government media frame diversity was correlated with public frame diversity significantly and positively (r = .32, p < .001), lending support for H3.

Dynamics of Networked Framing

H1b hypothesized that the magnitude of alignment between government media frames and public frames increased, and H2b predicted that their frame diversity decreased during the pandemic. We mapped the trends of frame alignment and frame diversity and tested our hypotheses using the Mann–Kendall trend test, a nonparameter test to analyze time series data for monotonic trends. In Figure 3, the alignment between government news frames and public frames became increasingly obvious and stable, while frame diversities of government media frame and public frame decreased over time. Results of the Mann–Kendall trend test showed a significant positive trend in the alignment of frames between media and the public (τ = .24, p < .01) and significant negative trends in diversity of media frames (τ = −.26, p < .01) and public frames (τ = −.48, p < .001), leading to the support of H1a and H1b (see SI P6 for details of trend analysis).

Discussion

We investigated the interrelationship and dynamics of public frames and government media frames on Weibo by conducting an automated frame analysis on big data. Our results show that the public frame network was less cohesive and more fragmented than that of the government media. There was a moderate level of frame alignment between the two actors: The public selectively redistributed certain government media frames or created their own to contest the official discourse. Government media frame diversity affects public frame diversity. The results are discussed in detail as follows.

Profiling Frames and Frame Functions in Government Media and the Public

First, our study showed that government media in China emphasized the role of government regulation in tackling the pandemic, as they primarily used the frames of “Public health outbreak” and “Government regulations.” This is consistent with findings from content analyses of frames used by government media in China (e.g., Ngai et al., 2020). “Domestic public health outbreak” was often connected with “Government regulations,” which became more dominant as the pandemic progressed. As the pandemic expanded globally, the “International public health outbreak” frame became more prevalent and appeared more often with other frames like “Consequences.”

Moreover, the government media’s frame network was relatively coherent, particularly in the pandemic’s later stages. This suggests that China’s government media managed to present a coherent official discourse to define the situation, propose official solutions, and perform value advocacy based on the interest of the party-state. Those official frames were relatively well connected, which probably enhanced the capacity of the Chinese government to define the issue and affect individuals’ issue interpretations. Together, these findings contribute to our understanding of networked framing a controlled environment by demonstrating how government media in an authoritarian society constructed and juxtaposed different frames through networked framing to strengthen their information control in an extraordinary situation (e.g., Gries et al., 2016).

In their reposts, the public often emphasized the “Human interest,” “Call to action,” and “Commemoration” frames, which highlight individual feelings, emotional support, and behavioral reactions that were largely absent in government media posts. Emotional expression was a prominent function of public frames, consistent with previous studies on the role of affective publics in networked framing (Papacharissi, 2016). These findings show how the networked public strategically paired various frames to express their emotions during the pandemic. The public not only refocused their crisis interpretations by selectively sharing media frames but also distributed alternative frames such as “Information transparency” and “Environmental protection.” For example, Weibo users held local officials accountable by questioning the transparency of pandemic-related information. The low frequency of public frames challenging official discourse may be explained by Weibo’s opinion censorship, which likely muddled the process of networked framing by amplifying selective exposure to official discourses and weakening collaborative filtering that might have pushed counterdiscourses to prominence.

Our results also indicate that the network of public frames was more fragmented than that of government media, especially at the earlier stage. As the pandemic worsened, individuals were likely more involved and motivated to engage with certain public frames. High fragmentation in the public sphere indicates few connections among different public frames. This might be explained by Weibo’s censorship of some high-profile pro-democracy events during the initial stage of the pandemic. Platform censorship likely decreased the number of counter-hegemonic frames and isolated them from existing prominent frames, which constrained their likelihood to be crowdsourced to prominence in the process of networked framing. These findings advance our understanding regarding how platform censorship conditions networked framing in a controlled environment (Nip & Fu, 2016b).

Frame Alignment, Diversity, and Dynamics

Next, we found that the level of frame alignment between government media and the public was moderate and increased over time. Initially, the pandemic was signaled by a discrepancy of frames between different actors, as the public disseminated a variety of frames to express their emotions, mobilize social support for crisis relief, and challenge the official discourse about the pandemic. As the pandemic dwindled, however, government media and the public both shifted attention to collectively coping with the crisis’s repercussions in various ways, such as by returning to work. As such, the frames of the two actors gradually converged as part of collective sensemaking. These results add to our understanding of the dynamics of networked framing in an authoritarian society (e.g., van der Meer et al., 2014).

Finally, our results suggest that the diversity of public frames was higher than that of government media frames, and their frame diversity decreased over time. These results advance our understanding of reframing (Chan, 2018) by showing how the diversity of public frames and government media frames shifts during a pandemic. Despite the tight control of the party-state, the public still engaged in generating and disseminating counterdiscourses to negotiate and contest the dominant discourse on Weibo. These results support the contextual and fluid nature of networked framing in an authoritarian society (e.g., Jiang et al., 2015). We also found that government media frame diversity was positively associated with public frame diversity. Given that public reposts temporally occurred after media posts, we may interpret this correlation as indicating that media frame diversity influenced public frame diversity, rather than vice versa, supporting past findings regarding media diversity’s beneficial effect on the plurality of public opinion (Huang, 2010).

Altogether, the relationship between government media and the public seems to have been rather complicated during an extraordinary situation like the pandemic. Government media strategically constructed a cohesive network of frames to enhance discursive control, which was further amplified by platform censorship to minimize the influence of counterdiscourses. The public challenged government media frames by deploying unique frame functions and creating new frames. Yet, those alternative frames existed only at the pandemic’s initial stage for certain topics that evaded censorship.

Theoretical Implications

This study made several important theoretical contributions. First, traditional framing research typically focuses on a discrete frame rather than multiple or even conflicting frames, compromising the external validity of the findings (e.g., Borah, 2011). Based on the notion of frame multiplicity, this study extends networked framing by illustrating how elite and non-elite actors constructed networks of frames and paired different frames in the discursive sphere to fulfill certain frame functions. Future research can investigate how individuals process and react to a network of frames in an experimental setting to understand the informational mechanisms and perceptual outcomes of networked framing.

Our findings also contribute to the literature on counterdiscourses and counterpublics (Chan, 2018) by demonstrating the mechanisms of networked reframing in an authoritarian society. The public engaged in reframing to define the controversial issues, but their counter-hegemonic counterdiscourses were not well connected to the existing frames. Future research should further investigate the role of censorship in networked framing and how counterpublics circumvent censorship to push their counterdiscourses to prominence. Future research can also examine a variety of counterdiscourses in different forms, like reposts, comments, and posts, to better understand the role of different affordances in counterdiscourse creation and dissemination.

Last, a dynamic perspective to networked framing captures the complex dynamics of different actors’ framing in a turbulent information environment. An examination of the shift of frame alignment and frame diversity helps us unpack the dialectical relationship between government media and the public. Future research should continue investigating the dynamics of networked framing among multiple actors in alternative social media and societal contexts.

Limitations and Directions for Future Research

Several caveats should be considered when interpreting our findings. First, public discussion can be eroded by heavy internet censorship. Although we excluded reposts from governmental and media accounts, public opinion on Weibo is liable to be contaminated by bot accounts, hired internet commentators, or the “50 Cent Party.” Future research, if possible, should rely on uncensored data as a less noisy measurement of public opinions. Future research can also consider employing innovative procedures to remove content by bot accounts or the “50 Cent Party” to capture public frames more precisely. Second, our study only focuses on government media as a unique kind of elite media in the Chinese media ecology. While our choice can be justified by the literature supporting the important roles of government media on public opinion during epidemics/pandemics (e.g., Ngai et al., 2020), future research should investigate the dynamics of frame co-construction by a variety of elite media like nongovernmental organizations (NGOs), government officials, or journalists.

Third, our findings could also be limited by applying topic modeling to Weibo data. Public discussion on social media is relatively brief, noisy, and equivocal compared with news content. Short texts contain less information, thereby limiting the accuracy of automated analysis. Besides, topic modeling discovers frames based on text similarity, while word ordering and topic dependence are ignored. Increasingly, communication scholars are turning to automatic approaches to deal with big data. How an automatic approach reaches high accuracy in prediction is crucial for achieving valid results. Thus, researchers should employ alternative methods to validate existing approaches of automated frame analysis (e.g., Wallach, 2006).

Conclusion

Our study has contributed to the literature by examining the dynamics of networked framing using big data. With a multifaceted theorization and measurement of networked framing, we found that the public contested official frames by distributing alternative frames and presenting frame functions divergent from government media. However, public counterdiscourse was less cohesive than an official discourse on Weibo. In a media environment featuring strict surveillance, social media serve as public forums posing minimal threats to an authoritarian regime. Our use of big data and computational methods enables us to reconsider traditional concepts like framing in the current media ecology and test new ideas about the interplay of media and public discourses. Future research can apply time series analyses to longitudinal data to further test the intertwined interactivity of media frames and public frames. Computational methods are still nascent, yet our study demonstrates how big data can be used to refine traditional communication theories.

Author Biographies

Xinyan Zhao (ezhao@unc.edu, University of Maryland, Ph.D.) is an assistant professor at the School of Journalism and Media, University of North Carolina at Chapel Hill. Her research focuses on the roles of social media and social networks in crisis and health communication using computational and quantitative methods.

Xiaohui Wang (xiaohui.wang@cityu.edu.hk, Nanyang Technological University, Ph.D.) is an assistant professor at the Department of Media and Communication, City University of Hong Kong. His research interests include information networks, health informatics, and social media.