2.1 SNs as an Arena for Public Opinion
2.1.2 Data-Sovereignty and E-Governance
2.1.3 Climate Change Discourse in China
2.1.4 New Physics of Information Dissemination
2.2 Network Architecture and Affordances
2.2.4 Juxtaposition and Behavioral Effects
2.3 Previous Research with Weibo
2.3.1 Elite Network in State-Regulated Weibo
2.3.2 Who Speaks for Climate Change in China – Evidence from Weibo
What do the protests of the Arab Spring[1], the hype about dogecoins[2], and Trump’s electoral victory[3] have in common? All three phenomena have inspired researchers to conduct a political analysis of social media. The diversity of disciplines is in contrast to the geographical monotony of the emerging research area. Searching for network analysis, sentiment analysis, or natural language processing on SNs leads almost exclusively to Facebook and Twitter. Regions where other platforms dominate are only sparsely covered. A practical reason for this is the inaccessibility of Developer API licenses for networks from such regions as well as the path dependency of analytics software already designed for the US juggernauts.
The research gap seems to be particularly large in the Chinese network community. With WeChat, Douyin, TikTok, QQ and Sina Weibo, five of the ten most-used SNs in the world are based in the PRC. [4] Ranked tenth, Weibo offers a 566 million MAU strong sphere of public discourse. The platform is particularly interesting for scholars of political communication, as content is more public than Tencent’s “inward-looking” messengers and less entertainment-centric than Bytedances’ short-video community.[5][6][7] In the winter of 2011/12, for example, Weibo became the medium for a campaign against Beijing’s air pollution. Under the hashtag “PM 2.5”, environmental organizations, bloggers, and private individuals exerted political pressure that led the government to issue new benchmarks and measurement metrics. [8] The role of SNs as a political space should therefore not be negated by a mere reference to the Chinese regime type.
On the other hand, however, the regime type should not be ignored as a context for observation. Behind the Great Firewall, an Internet with Chinese characteristics has grown.[9] Weibo is not the “Chinese Twitter,” but Weibo – a platform with its very own digital culture and network architecture. The fact that SNs have become an arena of the political is not only known amongst transnational activists and organizations, companies and governments also use the platforms to disseminate and collect information. According to an extrapolation based on a data leak, the Chinese government may have disseminated more than 400 million posts on Weibo in 2017 to mitigate extreme events capable of mobilization by using “distraction and cheerleading”.[10]
Political communication via SNs is in the area of conflict between civil participation and governmental public relations. Through network analysis, KOLs can be identified and put in relation to each other. For this purpose, metadata on posts and follower-followee[11] relations of Weibo are collected in a first step. However, to turn the raw data into politically relevant information, a delineation value is needed. A cross-platform comparison with Twitter’s network structures will serve as a context and benchmark for findings on information flow and opinion leadership.
Network analyses rely on the repeated appearance of individual actors, since it is especially those nodes, with multiple edges, which structure the otherwise unconnected clusters and thus allow conclusions to be drawn about thematic and institutional proximity. To create this reoccurrence, the analysis must be constrained by search terms. Climate change provides a good thematic limitation, as it is largely uncensored, discussed by all politically relevant actors, and encounters two network communities in a similar political contexts. Due to both the political and social nature of the topic area, it provides a starting point for a robust comparison.
Two levels of information flow are examined. An initial analysis provides an overview of the conversation participants and their connections. In a second step, the influential elite of this network is explored to answer the following questions:
Q1: How do the Network Structure of the Climate Debate on Twitter and Weibo compare?
Q2: Which Category of Actors Enjoys the Highest Network Centrality on each Platform?
Yet, the exploratory nature of the research method requires an openness to reflection and investigation based on the data outside of these predetermined questions.
First, a stable framework must be created through which the data will later be viewed. Analyzing different SNs from different countries of different regime types and different languages, there is a lot to be clarified. To develop a method, the following points must be addressed. First, an understanding of the political nature of social media must be developed. Since the aforementioned geographical bias has already led to extensive research on the American and European cyberspaces, special attention will be paid to the largely neglected context of political communication in China. By looking at the current catchphrases “data sovereignty” and “e-governance”, the relationship between China’s political elite and the emerging net communities will be illustrated. A review of China’s climate debate will provide some context and show why the topic is suitable for an examination of political communication. Then, the new logic of information dissemination will be analyzed in contrast to traditional media. Thereafter, the insights are applied to the two platforms. Here, Twitter and its climate debate serve merely as a reference point for the analysis of Weibo. In the search for data points that are not distorted by the differences in network architectures, a systematic comparison of the digital context of origin will be conducted. With this knowledge, a method of data collection and processing is created. The findings are presented together with their visualizations and provide a starting point for the research of political communication within the climate discourse on Twitter and Weibo.
At 4.55 billion people, 75% of the world’s population over the age of thirteen used at least one SN in January 2022. On average, 150 minutes are spent on seven platforms per day, 25 minutes more than on digital and analog newspapers. 35% of users list reading news among their main activities, 29% want to know what others are talking about, and 20% cite contact with journalists, experts, and influencers as one of their primary motives. [12] Of the ten most-used SNs in the US in 2021, Twitter is the one where most users consumed news on the platform at least sometimes, at 55%.[13] Although it is one of the largely commercialized “Super-Apps” that the Chinese cyberspace is famous for, Weibo is the most open and political option for our purpose.[14]
The fact that social media has become a crucial element in the public information flow and thus an arena of public opinion is hardly debatable. But what exactly this influence looks like – and who benefits from it – certainly is.[15] Many scientific papers with network analyses based on data collected from SNs distinguish between a “more concentrated and hierarchical” and thus authoritarian and a “more dispersed and egalitarian” democratic informational deliberation.[16] But caution must be taken not to stray into the well-trodden paths of politically distorted speculation. In the search for characteristics of political communication the focus must be on the analysis of the internal dynamics of information dissemination process.
” Therefore, to better understand China’s climate politics, we have to move beyond a dualistic view that rests on a binary opposition between state and civil society, and direct more attention to the processes through which state and civil society interact, as well as their contexts and dynamics. ” [17]
Although all-encompassing theories about the relationship between political communication and regime types are avoided, it is rewarding to look at the political context in which the data were collected.
In line with its long-term goal of becoming more independent of international trade, Beijing has developed profound programs to strengthen digital sovereignty. Elevated to the level of national security, the theme aims to minimize and regulate cross-border data flows, reduce exposure to sanctions, minimize foreign surveillance, while fostering domestic surveillance and protecting the huge Chinese market for domestic Internet companies. [18][19]
The endeavor runs hand in hand with another characteristic of the Chinese governmentality. The early recognition of the importance of data moved the government itself into the digital realm. Among many other forms of appearance, Beijing’s digital ambition manifests itself on SNs. The recent crackdown on Chinese internet giants also hit Weibo. From January to November 2021, it was fined 44 times totaling 14.3 million RNB.[20] Officially, it’s about data breaches, but behind the scenes, Beijing is talking about spiritual pollution and underlines that the cyberspace is a part of its sphere of influence.[21] Yet Beijing has more tools at its disposal than just sanctions. For example, the CCP presented Xuéxí Qiángguó [学习强国][22], its own SN with more than 100 million MAUs in 2019.[23] But Beijing also makes use of established SNs. The CCP not only experiments as a proactive intermediary but only issues licenses only to those companies that comply with Chinese law.[24] Accordingly, Weibo is not allowed to contribute to the dissemination of the following:
This is not an exclusive feature of the Chinese cyberspace. Twitter and Facebook have also been identified as political actors and have been restricted by law. Regulations require the removal of bots, the prevention of the spread of false news, and the blocking of accounts with illegal content. Although Elon Musk promises upcoming deregulation following the acquisition of Twitter in 2022, there has been a trend toward juridification of the digital space among the American internet giants since the scandal surrounding Cambridge Analytica at the latest.
These developments lead to two suggestions for further research. First, scholars should primarily work with official APIs, since web scraping, as practiced in scientific papers in the past, has now become tough. Secondly, special attention should be paid to the various types of government influence.
Of course, not every topic can be covered in a cross-platform comparison. It is necessary to isolate the network structures as the object of investigation. To do this, as many disturbance variables as possible must be equalized. But why is the debate about climate change particularly resistant to those external distortions?
Identical Distribution of Roles: A multitude of issues would be accompanied by the problem that the two states under consideration take different positions. For example, in the case of the Beijing Winter Olympics 2022, the Hong Kong protests, and the trade conflicts, the Chinese government takes on a different role than its US-American counterpart. All outcomes would thus not only be an indication of political communication, but simultaneously, and more importantly, to an incalculable extent, also a testimony of correlations with variables such as hosting or taking part in the contest, political initiative or reaction and geographic proximity. The climate change discourse, on the other hand, offers some useful commonalities. The two dominant states behind the English-speaking Twitter and Weibo are equally confronted with the threats and opportunities of climate change, share political responsibility and firepower, and are largely adherent to the scientific climate consensus.
Public Discourse: After the 2009 climate conference in Copenhagen, conspiracy myths boomed in the PRC. Many authors felt that “the West leveraged its scientific authority to impose restrictive policies on China”.[26] Only two years later, President Hú Jǐntāo promised to play a shaping role globally and “seize preemptive opportunities in the new round of the global energy revolution.” Unprecedented investment in renewable energy installation and the wording of an ecological civilization (生态文明) became a stringent narrative: “Climate change no longer made China look weak. It was now a story of China’s strength.”[27][28] It was not only the political elite that joined the global climate consensus. In the years to follow, NGOs, academic communities and a public debate developed around the issue.[29][30] It can therefore be assumed that not only do all politically relevant actors participate in the discourse but that they also interact with each other in a way that offers transferability to political communication in general.
Keyword Searchability and Censorship: The debate surrounding climate change is mature and widely equipped with well aligned argumentation patterns and internationally agreed interpretative paradigms, terminology, and symbols. Discourse often hides behind its own subcategories. Data on nationalism, for example, cannot be collected by searching for the term itself. The variable is hidden behind loose concepts, which in turn may themselves be part of debates that have no connection to nationalism. The concepts of climate change (气候变化) and global warming (地球变暖), on the other hand, play a central role in the debate. Neither are they frequently used in posts that do not actually seek to engage in this discourse, nor is there a large proportion of climate change-related posts in which these words are not used. Both platforms have a liberal attitude towards the topic. A crosscheck on FreeWeibo shows that no post with either keyword was removed in the first three months of 2022.
To determine which data is important under the new logic of the modern media landscape, it is necessary to analyze the physical properties of the information flow of social media.
With some interactive exceptions, newspapers, radio, and television offer a one-dimensional flow of information. The longstanding role distribution of news-creating agencies, processing media and merely consuming end-users has recently been suspended. Advances in ICT have largely eliminated technical infrastructure as the bottleneck of mass communication, enabling nearly five billion Internet users to emerge as potential broadcasters.[31]
“ The process by which people locate, organize, and coordinate groups of individuals with similar interests, the number and nature of information and news sources available, and the ability to solicit and share opinions and ideas across various topics have all undergone dramatic change with the rise of social media. ” [32]
The physical characteristics of the information flow of the “new media” or “digital media” can be grasped in contrast to the once-dominant media through the following lenses:
The Direction of Information Flow: SNs paved the way for oncoming traffic on the formerly unidirectional road. On the one hand, this means that now a majority of the actors involved can also assume the role of creators and disseminators. On the other hand, the previously merely sending minority had to open up to digital participation. Instead of the strictly vertical direction of information flow, we thus observe a less structured and horizontal “versatility and dynamism of relationships” in political communication on SNs. [33][34] This description, however, is more in the realm of potentialities. One should refrain from viewing SNs as “intrinsically flat, egalitarian and grassroots-centered,” because here, too, one finds the “hierarchical social field based on an uneven distribution of digital capital” familiar from traditional media. [35][36]
Disintermediation: The need for intermediaries in the chain of information diffusion disappears. Senders can distribute themselves and receivers can find their own sources. The elimination of these often-regulated nodes decentralizes public debate and creates a space for subcultures and political niches, which can isolate themselves in “highly segmented and homogeneous” echo chambers. [37][38]
Gate Keeping und Agenda Setting: The weakening of this intermediary role also results in less effective regulation of contents’ visibility. Hubs of communication moderation can also be found on SNs, but even spontaneous clusters of otherwise dispersed users can trigger “independent trends without diffusion through external entities”.[39][40]
Organization und Mobilization: SNs reduce the financial and time-related transaction costs of political communication. Users are divided into dynamic cohorts as “prosuments” by the platform-specific algorithms. The separation and its associated network-effects stimulate the collective construction of narratives, argumentation patterns and associations. The platforms become the vehicle of politically charged symbols, concepts, and emotions. As emotional states of heightened physiological arousal, negative emotions such as anger or fear are catalysts for social diffusion and dissemination.[41] Movement can thus exert political pressure even without institutionalized support.
Datafication and Programmability: Like any other medium, social media has a distinctive logic within which all actors must navigate. Like the ongoing process of commoditization, in the age of dataism, metadata previously considered a useless byproduct must be quantified, aggregated, and made meaningful in the web of correlations.[42] The process of making behavior, sentiment and social interaction numerable, penetrates the concepts of both network architecture and user behavior.[43] The entire surface of SNs can be seen as a toolbox for quantification. The limited selection of possible interactions, anchors users in a designated behavioral corridor.
Even in the age of social media, the question of the visibility and reach of information is not independent of factors outside the information itself. These factors need to be systematically investigated and comparatively classified, to answer who has taken the lead in replacing the former monopolists of agenda-setting and gatekeeping.
The user’s programmability cannot be undone or factored out by analyzing it. However, an analysis should point out the biggest sources of error in potential comparisons and sharpen the critical sense for dealing with political communication on SNs. In a first step, therefore, it is not the data but the network architecture of the platforms in question that is classified in terms of their function as vehicles of political communication.
Any survey must be embedded in a methodological framework. When creating the design, possible biases must be identified at an early stage and counteracted by randomization, weightings, filters, and many other methods. An entire field of research with a lush vocabulary for psychological effects and potential fallacies has emerged around the creation and implementation of surveys. Although SN datasets often boast numbers of participants that are virtually unattainable in physical surveys, they still cannot escape critical scrutiny. User behavior should by no means be approached with the notion of a mystified authenticity. If it exists at all, it should not be sought on SNs, where every user knows about the panoptical observation by a broad public and the platform itself. The still embryonic field of “Human-Computer Interaction” seeks to define an adequate approach to social media data and already has illustrated some pitfalls.
A comparison of comments on the largely anonymous YouTube and the more personalized Facebook shows a difference in politeness.[44] This result is neither surprising nor does it invalidate the data. However, it is a warning for the possible errors of cross-platform comparison. If it wasn’t the relationship between anonymity and politeness that had been studied, but common patterns of argumentation or emotional attitudes of two groups of people which are unequally distributed across the platforms, an almost incalculable bias would have emerged. Often, the variations in network architecture are far more subtle than the dichotomous distinction of anonymity. The layout of the user interface and the number of clicks required to reach the target have a significant influence on digital behavior. Actions aren’t determined but suggested by “sociotechnical affordances”, which are defined by reachability and snugness. [45][46] But it gets even more confusing: Beyond the syntax of the platforms, specific cultures of interaction are established.[47][48] Why, for example, are posts on Instagram liked much more frivolously than those on Facebook? The fact that Instagram presents the entire post as a hitbox of the like seems to explain only part of the difference, given the similarily high accessibility of fakebooks “thumbs up”. Rather, the observation shows that the value of any interactions is newly negotiated on every platform.
“ Lisa Gitelman aptly coined the adage “‘raw data’ is an oxymoron”, meaning that data are always already prefigured through a platform’s gathering mechanisms. Moreover, in processing data, a platform does not merely “measure” certain expressions or opinions, but also helps mold them. In opening up “spontaneous” sentiments and opinions to the public eye, platforms have rendered them formalized and preformatted expressions—even though many tweets appear, to say the least, unpolished. […] The idea that you can tap into people’s unconsciousness or “idea formation” without affecting the processes of opinion making is a basic misconception, which goes back to the classic observer effect—a concept familiar to research method literature across disciplines. ” [49]
To avoid a hasty decontextualization of “raw data”, the network architecture and it’s expected effect on user behavior must first be analyzed comparatively. Since both platforms are mostly accessed via mobile devices and considering that the desktop version is governed by the same network rationale, we will focus on the mobile apps.
With 436 million MAUs[50] and about 200 billion tweets per year[51], Twitter has become a benchmark of open public communication. Twitter describes itself as a “global platform for public self-expression and conversation in real-time”. The platform “allows people to consume, create, distribute and discover content and has democratized content creation and distribution”.[52] But how exactly does Twitter translate its stated goals into its network architecture? To lay a foundation for a stable comparison with data from Weibo, both platforms are examined in terms of their registration process, user interface, interactions and content mediation.
Registration Process and Anonymity: To register, Twitter only requires an email address to receive an authentication code. Unique alphanumerical pseudonyms can be freely chosen and many of the features to personalize the feed (IP address, Device, Demographic Data) are in effect by default but can be disabled.[53] Since single-use email addresses are sufficient and account settings allow private voids that do not extend beyond the specifically selected contacts, Twitter can potentially be used anonymously. However, Twitters’ collective customs and personalized default settings convince 68% of users to use their real names, with just 6% remaining entirely unidentifiable.[54]
Graphical User Interface: “The GUI dictates the look of the social medium’s home page, how a user navigates across different spaces within the platform.” [55] Twitter’s GUI places a centralized home feed for aggregating, ranking, and displaying content at the core. This is embedded by a less prominent search bar, that unfolds a personalized trend feed with frequently used keywords, as well as some navigators to one’s profile, personal messages, notifications, and settings. The feed expands endlessly by scrolling, interrupted by recommendations for new hashtags, accounts, and theme conglomerates called “lists”. Stylistically, the look is clean, intuitive, and structured. A toggle to switch from the algorithm-based “top tweets” to a purely chronological feed is unobtrusive, and even though the dropdown menu doesn’t display any category that isn’t already represented on the home feed, a second click is required to confirm the switch. A salient blue button to post your own tweet follows along almost all tabs.
Interactions: Profiles are public by default and networking is uni-directional.[56] This means that there is no reciprocity of post visibility.[57] Visibility, however, can be enforced through private direct messages. On the home feed, Twitter offers a total of four ways to interact with Tweets. They can be liked, retweeted, quoted and commented on. The comment function has a surprisingly peripheral position. The button below the posts only allows users to write their own comments. If they want to read other users’ comments, they must click on the post and find what they are looking for in a tab that is sparse from the home feed. When authoring tweets, hashtags can be used to gain access to a spontaneous and diffuse debate beyond one’s designated audience.[58] Hashtags experiences searchability in a separate feed, especially on the desktop version.
Algorithmic Content Mediation and KOL-Attraction: Algorithms mediate the selection, sequence, and visibility of posts by promoting and limiting their reach.[59] Algorithms are at the heart of any SN and subject of constant refinement. In a blog post from 2017, three factors are presented. First, the Tweet itself is examined: What counts is its recency, whether it contains multimedia and indexed hashtags, and how many interactions it has generated. The author also plays a role: Here, the history of interactions with previous posts and the personal and thematic overlap are decisive. Finally, the user must be considered: Metadata on user behavior and the interaction history are of importance for this last factor.[60] According to the logic of datafication discussed earlier, every interaction is of value. They are the basis of psychographic profiling and can thus contribute to the improvement of ever more nuanced microtargeting. Companies, organizations, and individuals pay Twitter to momentarily refrain from maximizing the collection of data points by granting their post disproportionate reach. Twitter not only filters and delegates content but also indirectly contributes to the creation of content through the payment of designated KOLs.[61]
With 566 million MAUs[62] and more than 30 billion posts per year[63], Weibo has a prominent position in the Chinese media landscape. The platform offers no social self-portrayal comparable to Twitters. The annual reports are devoted to quantitative and legal matters only, and the customer service lists the dissemination of news as the third point after contact with friends and celebrities, and immediately puts it into perspective: “Of course, in addition to the news content, there is also more entertaining information.”[64] Nevertheless, Weibo plays a similar role in Chinese discourse as Twitter does in America and Europe.
“ Like Twitter, Weibo was designed as a public platform with information-centered communication and its core function emphasizes the production of information: Enabling citizens to become the source of information. ” [65]
“Chinese Twitter” may be the answer to the question of which two words can be used to evoke the most precise presentation of the platform outside of China. However, anyone who has a few more words to say about it should refrain from using this analogy. “Sina Weibo was launched in August 2009 to fill the void after Twitter was blocked by the Chinese government since the Ürümqi riots that July.”[66] At first, Weibo (微博- Microblog) was a systematic replica of Twitter, but later on, they adapted “by adding unique features originating from Chinese Internet culture, such as rich media, threaded comment, private chat, microgroup, and microevent”.[67]
Registration Process and Anonymity: Since 2017, users must provide their real names to be able to write posts themselves. This is controlled by capturing a mobile number. In China – as in most countries around the world – these can only be activated by submitting an official identification document. However, the legal name does not have to be the same as the username. The corresponding legal text states: “Real names behind the scenes, voluntary on stage.” [68]Unfortunately, no official figures could be found on this, but an unsystematic user sample, as well as some forum posts about anonymity on Weibo, show that the offer for pseudonymization is widely used. Although “Sina uses its recently introduced VIP system to encourage all users not to communicate through online names but to be recognized by their real offline identities” this feature is almost exclusively used by celebrities, organizations, companies, and political actors.[69]
Graphical User Interface: Weibo’s GUI is clearly structured differently. Instead of nine icons, of which three slip out of the screen when scrolling, Weibo offers 13. Weibo expands the already mentioned buttons with features for short videos, gamification, recording pictures directly via the app, and a largely separate area for centralized trends. The overall image is less orderly and structured. [70] Comments are not in thread format under the videos but float along the top of the screen unasked. A mixture of games and verifications, a fan-shop, and the Weibo-Wallet point to the culture of superapps.
Interactions: The spectrum of possible interactions is largely equivalent to Twitter. Instead of quotations, Weibo enables “fast repost”. Hashtags play a subordinate role since the “discover-feed” does not depend on them. Frequently used words are indexed even without hashtags. A self-classification into thematic categories and keywords by the usage of hashtags is therefore much less widespread.
Algorithmic Content Mediation and KOL-Attraction: Weibo’s home-feed is based on the same three multiplicative factors as the EdgeRank-Algorithm once used by Facebook. The first is intimacy, which is measured by the frequency of interactions. Second is the quality score, which is measured by objective metadata, such as the number of characters, the use of headlines, and multimedia as well as the authors’ interaction record. The last factor is a simple time-based decay parameter.[71] The Sina Corporation also intervenes in the content creation through KOL attraction. To do so, they equip content creators “with the opportunity to monetize their social assets on Weibo through advertising, e-commerce, paid subscription, tipping, and other means.”[72]
Category | Fields of Expected Behavioral Effect | ||
Registration Requirements and Anonymity | – Email address – High anonymity potential – Low pseudonymization | – Mobile number – Low anonymity potential – High pseudonymization | – Sentiment Expression – Subversion/ Opposition (esp. unlawful content) |
Algorithmic Content Moderation | – PageRang (Post-, Author-, User-Characteristics) | – EdgeRank (Intimacy, Quality, Recency) | – Largely intransparent |
Graphical User Interface
|
– Simplistic and clean – Prominence of Home-feed
| – Convoluted and messy – Multimedia dominance – Superapp (Gamification, Minivideo, Fanshop) | – Usage Motivation and Content (Commercial – Informational – Social – Entertainment) – Commercial Activity |
Social Interactions
| – Peripheral comment section | – Indexing without hashtag – Reference by hashtag rather than @Mentioning | – Thematic Proximity and Content Overlaps – Comment Analysis |
This table does not claim to be exhaustive. Rather, it is meant to give an idea of how the (different) network architectures can sabotage research results. Especially those data that are destined to become part of a comparison must be handled with extreme caution. A cross-platform comparison of political opposition would probably already be distorted to the point of uselessness due to different anonymity distributions, a comparison of thematic overlap through hashtag networks fails due to the independent indexing of Weibo, and a content-based comparison of usage motivation and posted content describes the users and the digital niche of differently oriented platforms in the same breath. However, the data points of social interaction needed for user network analysis appear to be only marginally affected. Apart from comments that are distorted by differences of the GUIs[73], Reposts, @Mentionings and quotations perform comparable tasks and have similarly prominent positions in the respective network architecture. In addition to these considerations, a look at the methodological approaches used in similar research will help.
For network analyses with data from Twitter, one of the many predefined paths can be followed. Theoretical frameworks are developed, data collection and analysis are illustrated in numerous tutorials, and findings can be interpreted in a sophisticated web of reference values. [74] Unfortunately, there is no comparable body of knowledge for Weibo. Therefore, it is worth looking at the small selection of research projects with special caution.
Lin, Hamm, and Reinhardt also investigated political capital within the Chinese online community in their 2018 paper. They similarly perform a network analysis of Weibo’s KOLs. Therefore, “different starting points from all channel categories are chosen” based on literature and media reviews.[75] From each account, the entirety of the followees and the number of posts are taken and fed into Gephi’s data lab. They identify a total of 291 channels, 43 percent of which belong to the entertainment and arts category, 45 are media (not further differentiated), 39 are Weibo services and 13 are governmental.[76]
The resulting graph can be divided into three divisions: The entertainment-centered, artistic community, a distinct and well-connected media cluster, which is found in isolated cases throughout the network, and finally a cluster with Weibo services. In a second graph, the same actors are classified according to their political interests. Two-thirds are said to have an apolitical agenda, and of the remaining 105 actors, “34 are coded to display culture politics, whereas 71 contain power politics”.[77] The reason for this is the commercial nature of the platform, which encourages elite accounts to keep away from political topics and instead focus on “lighthearted content” in consideration of their “fan economy”.[78]
A final distinction defines a vast but less active cluster of “escapist entertainment” that uses existing rules without challenging them. The establishment, dominated by the media and GONGOs, upholds the narratives of the CCP. But there is also a state-critical division. They question and re-contextualize political decisions and their creators within a well-defined corridor. The question of whether social media can be understood as a grassroots catalyst or as a propaganda machine of the powerful is met with a multi-layered complication. On the one hand, the CCP would paternalistically intervene in Chinese cyberspace and thus oppose the idea of Internet autonomy. On the other hand, the booming social networking technology in China today creates space for bottom-up activism.
Lin, Hamm and Reinhardt’s valuable research demonstrates the broadest structures of Weibo’s digital elite. Apart from the fact that the data collection would no longer be methodically possible anyway due to new restrictions, two promising alterations have been identified. Firstly, the dichotomous data points from the followee list (friend or not friend) lack informational depth. Social interactions that can be carried out more frequently (like, comment, share, @mentioning, quotation), on the other hand, offer the possibility of differentiating intensities of affiliation. The supposed importance of Weibo-Services would quickly disappear through this distinction and paint a more accurate picture. Secondly, the endeavor could benefit from a thematic restriction. This could facilitate the structure-giving repetition of actors and provide some more transferability due to its specificity.
Chung En-Liu and Zhao also wanted to counter the geographic bias of the research field. Using a web scraper specialized on Weibo, they collected 98 thousand posts with the keywords “Climate Change” and “Global Warming” in the two months before, during, and after the Paris Climate Summit of 2015. They have been published by 47 thousand accounts, of which the most shared one percent makes up more than 90% of all reposts, while more than 80 percent are not reposted even once.[79] The top 20 is constituted of 11 media companies (5 of them government-owned), 5 individuals, and 4 international organizations, nevertheless the three most shared posts are all from public figures from the arts and entertainment industry.
They conclude, that “Climate Change remains ‘safe’ as long as it appears to be a global and disembodied subject matter”. Therefore, the results are only partly applicable to other topics. The debate about air pollution in coastal metropolises threatened to mobilize widespread protest and “thus reporting need[ed] to be minimized and regulated”.[80] Finally, they oppose the notion of a civic-deliberative nature of Weibo and other SNs. Those platforms are “still constraint by the larger social structure” and merely repaint, not revolutionize, the already familiar power relations of information dissemination.[81]
With these results, Chung and Zhao have already made an important contribution to understanding the Chinese climate change debate. However, the findings need to be expanded in two ways. Firstly, a more detailed categorization is desirable. In particular, the two largest categories, “individual” and “government-owned” should be broken down in more detail to gain a deeper insight into the political capital of Weibo. Secondly, some of the results hover in empty space. Although the significance of demographic, geographic, or categorical characteristics can be identified, the proportions themselves lack a reference value. For example, it is clear that governmental actors are represented more prominently than media actors. However, the statement that governmental actors are strongly represented on Weibo requires differentiation from a comparable network. Therefore, we conduct research using Twitter as a reference.
Twitter provides all its users with the opportunity to apply for a Developer API license. There is no authentication of the submitted identity. Only a brief description of the method and the intended mode of publication are required. Just a few hours after the successful application, four credentials (API Key, API Secret, Access Token, Access Token Secret) are issued, which must then be incorporated into programs and tools. These tokens are most often embedded in the already well-tested and technically sophisticated Python universe. However, since the research question cannot be answered with the posted content, but only with the relationships of the actors, rudimentary scrapes of metadata via Gephi are sufficient. The open-source visualizer offers the possibility to collect data about user-relationships, hashtags, and emojis through the plug-in “Twitter Streaming Importer”. Although users could also be grouped according to hashtags and emojis, the identified differences in network architectures and common usage are too great for comparison. The scrapes therefore only consider @mentions, retweets, and quotations. In addition to the two search terms, the English language is set as a further limitation for more comparability with the largely linguistically homogeneous Weibo community.
Gephi instantaneously generates an initially random arrangement from the collected nodes and their connections. Only in a further step, the components are structured in a meaningful relationship by an algorithm [see section 3.3]. In an associated data lab, nodes and edges are listed with some metadata. In addition to a timestamp, data points on the number of followers, followees, and tweets can be found here. In the section on edges, a distinction is made between three possible types of connections: Retweets, @mentionings, or quotations lead to edges, which are not further differentiated in the following. It can probably be said that retweets tend to convey an implicit expression of liking, while quotations can just as well be used for an open confrontation. However, the features are not used discriminately, and a small sample shows that all three types of social interaction can contribute to meaningful network analysis. However, for smaller data sets, caution should be taken to avoid the possibility that actors whose exchanges consist of mutual dissociation may be mistakenly placed side by side. For example, the cluster of climate skeptics trumped the number of connections to Joe Biden in a midterm result. The law of large numbers, however, ultimately ensured the usual homophily effects of social media.
Since only live tweets can be collected via Gephi, it is not possible to collect the feed from a particular event in global climate policy. However, since climate change is also widely debated independent of such events and the periods between summits and catastrophes are politically significant as well, this circumstance is acceptable. To have some resistance to day-specific outliers, data were collected permanently for the first ten days of March 2022. Despite minor interruptions due to technical difficulties, 265 thousand nodes with 684 thousand edges were collected during this period.
Weibo also offers an API, but the application process is complicated and time-consuming. The very fact that no spaces are accepted in the forms spoils the outcome of a first application attempt: “Reason for rejection: Sorry, overseas users could not pass the examination.”[82] A colleague provided the required Chinese mobile number and identity card, but the subsequent three-step authentication demanded programming proficiency that was not feasible given the sparse tutorial situation on the subject on Baidu.
Other programming solutions are also limited by Weibo’s security efforts. Automated scraping with Python’s BeautifulSoup ends up back in the login. Responsible for this is probably an invisible CAPTCHA, which is triggered by the lack of a shaky cursor or the high click frequency. Reading the HAR protocols which capture the communication between the browser and Weibo Servers leads to almost unintelligible piles of data due to the bulky representation of Chinese characters and the many features beyond the targeted feed. With a lot of RAM, CPU, and expertise in handling pythons R-Selenium, Pandas, and RegEx, scrapes of the HTML-texts or the HAR-files could be made possible, but Weibo poses a completely different kind of problem. Some data points are denied to users on a very basic level. For example, only the first 200 followees of each account can be viewed. The analysis of the elite-network presented earlier, would therefore not be possible today. Also, a list of accounts that reposted something can only be taken from the home feed. For search queries, the navigator loses this function and is limited to the number of reposts. In the advanced search settings, search terms and time can be specified. However, the results only show those actors who use the specified keywords themselves. Uncommented reposts of climate change posts, therefore, do not appear in the dataset that will be dealt with later. The number of posts thus remains a fraction of the Twitter Scrape. But since there is no apparent reason to suspect that certain clusters or categories of actors make particularly heavy use of a certain form of sharing, even this smaller dataset can yield robust results. The first ten days of march entail a total of 4134 posts with at least one of the keywords. 2078 (4134 minus overlap and those without any relational expression) actors and 3046 links are extracted from them.
The next step is to find a way to process data from SNs in accordance with the research goals. As in almost all contemporary computations of network analyses, the Force 2 algorithm is used for the subsequent graphs. Within it “Nodes repulse each other like charged particles, while edges attract their nodes, like springs. These forces create a movement that converges to a balanced state. This final configuration is expected to help the interpretation of the data.”[83] All 4 Graphs are structured according to the following rules:
This definition has been adopted from Lin, Hamm, and Reinhardt. [84]
Thereby, informational communities can be distinguished, and their most important actors pinpointed. A central position in the network goes along with the mediating power over the information flow. It therefore translates directly into “political capital” and the “capacity of political communication”.[85]
The research design was inspired by the papers discussed earlier and the methodological overview on the analysis of political communication on social media by Stieglitz and Linh. While a keyword-based approach was chosen for the data collection, an actor-based analysis will be applied now to identify the most central actors in the climate debate.[86]
“ There are a number of different measures of influence of an actor in a network. Basically, influence is determined by many factors, such as the novelty and resonance of their messages with those of their audience and the quality and frequency of the content they generate (Romero et al. 2011). In particular, resonance in terms of retweets (Twitter) and comments (Facebook and blogs) can be modeled as edges connecting nodes that represent users within a social network. In this way, different metrics of the concept of centrality and prestige can be applied to measure influence (e.g., degree, betweenness or eigenvector centrality; degree, proximity, or rank prestige). ”
Each of the listed metrics of centrality has its raison d’être. But they all represent only one dimension of political capital. The degree centrality counts the number of edges of each node and reveals “who can call upon the most resources”. The betweenness counts how often a node is on the shortest connection of other nodes and measures the “capacity to interrupt the flow of information”. The closeness measures the average distance in nodal jumps and thus defines those nodes that “could most quickly spread information”.[87] You will find a graph with these measures behind the QR-Code at the end of the paper.
However, the following illustrations are based on the centrality metric “eigenvector centrality” or “prestige centrality”. It’s distinctive feature is that not only neighboring nodes but the entire network is taken into account to determine the centrality. This way nodes with a “wide-reaching influence” on the “macro-scale” can be determined.[88] Nodes that have links to nodes with many edges are of greater relevance than those that are interlinked with nodes that are otherwise hardly interwoven. The same experience can be drawn from the information flow of SNs. Links by a few well-connected actors can generate more visibility than thousands of barely connected actors.
The desired classification into categories is not to be found in the scraped data. Since this work can only be done manually, the number of nodes must first be reduced. A limit of 200 nodes seemed appropriate given the edge density and the proportion relative to the smaller Weibo-network. Actors were classified into eight categories based on their profile name, profile picture, description, number of followers, and first 20 posts. The classification is largely adopted from Lin, Hamm, and Reinhardt, although some modifications were made to reflect the thematic limitation to climate change.[89]
In addition, three dichotomous variables were collected from all user categories. The distinction between governmental and non-governmental is needed as a refinement of the state actors category since it does not cover state-organized Institutions and media. The dummy variable of climate denial forms the only content-related item to be considered. It can be seen as an indicator for informational access participation of subversive communities. Finally, the distinction of verified channels can give a more precise idea of how central the platform itself is in the dissemination of information.
After the presentation of the two overview-graphs [4.1 and 4.2], these categories are to be found in graphs [4.3 and 4.4] as well as in table [4.5].
To answer Q1, we will first look at a pre-sorted network of all nodes of Twitter. It is intended to provide an overview that can be used to identify and relate clusters. The identified network structures will then be used as a benchmark to analyze the Weibo network.
Since a presentation of 265 thousand nodes and 684 thousand edges is neither technically feasible nor meaningful from an informational point of view, the 10 thousand nodes with the highest Eigenvector centrality were filtered out. To prevent too many color-differentiated clusters, the community detection algorithm was performed at high resolution (70). Clusters were emphasized using Lin-Log Mode (Linear Attraction & Logarithmic Repulsion).
It is a common and justified criticism that network analyses that flaunt too many nodes are a spectacle of an aesthetic rather than informative nature. Graph 1 is nevertheless intended to provide an overview outlining the complexity of the climate debate. The graphic is first analyzed as an isolated research object and put into comparison in a second step. Fundamentally, the following characteristics can be identified:
Solid Core with two Peripheral Clusters: Since the modularity resolution is chosen arbitrarily, it makes no sense to bother with quantitative descriptions of the constellation. Nevertheless, Gephi’s clusters allow an intuitive three-way division of the debate.
In the center, there is a dominant, Yellow Cluster. It includes the best-known international organizations (UN, IPCC, COP26, UN Woman, UNDRR, WRI Climate) and all those activists and bloggers who regularly refer to them (Greta Thunberg, GoGreen, Climate Now, Carbon Brief, Climate Nexus). A large part of the academic community is also in this cluster (Bill McKibben, Allan Margolin, Colin J. Carlson, Bradly Dennis, Michael E. Mann, Saleemul Huq). What’s striking is the prominent function that these scientists occupy. Although follower counts are far behind those of influencers located in the same cluster, they particularly often serve as connectors between otherwise isolated clusters. Another notable feature is the clear visibility of geographic distance. The lower reaches of this cluster are almost exclusively outside the transatlantic region. On the left is an African-dominated part, next to it a less pronounced Asian one.
The next largest Blue Cluster is home to almost as many nodes. However, their arrangement has a different texture. Instead of highly condensed subclusters, the network spreads out widely and evenly. While the yellow cluster is fragmented by a few local heavyweights, the blue cluster is characterized by continuous transitions and the absence of dominant nodes. Fewer organizations and academics hide behind the nodes. Bloggers, activists, political commentators, and public figures from the entertainment industry and other celebrities dominate here. In addition, this cluster can be said to have a political tendency. The cluster can be classified as belonging to the left political spectrum. Of the twenty largest nodes, 13 contain at least one term from the word families of “capitalism” (problematized), “socialism” or “feminism” in their description. Since manual categorization is very time-consuming, systematic analysis is only realistic for a smaller number of nodes, so we must rely on the vague trends of small samples.
The Orange Cluster on the left side is far less homogeneous. In its left periphery, there is a dense assembly of highly interconnected nodes. Further up the graph, isolated dominant players are found in a largely unpopulated terrain. With several exceptions, the factions can be classified as moderate and alternative right-wing. Going further up the graph means sliding even further to the right in the political spectrum. For the 20 most central actors, no political dogmas comparable to the blue cluster can be found in the actor’s description.
From “Show me the evidence that wind energy works…. still waiting” to accounts calling themselves “invaders of anthropogenic climate change denier echo chambers”, a diverse variety of political motivations can be identified. If there is a unifying criterion at all, it would be the contestation of the status quo. In this context, climate change often plays only a subordinate role in a general opposition to the political mainstream. Whether the high number of climate skeptics among the most influential accounts within this cluster is merely evidence of their good networking, or whether in fact a significant proportion of the Twitter community has some openness to this opinion cannot be determined in the absence of a content analysis. What can be said, however, is that of four centrality metrics and two data points from the accounts themselves, only the number of followers did not identify a double-digit percentage of denialists among their respective 200 most central actors.
Unfamiliarity of Central Actors: If you stare patiently enough at the high-resolution PNG, you will find the supposedly “big names” of the climate debate. In this analysis, however, they lack their perceived prominence. Scientists with access to the public and bloggers with a focus other than environmental policy benefit from the method of measurement. Their role as an information pathway between media companies and organizations, or as the only political access point for otherwise apolitical clusters, gives them a powerful exclusivity. Another explanation for the counterintuitive results is offered by the symbiosis with Twitter’s network architecture. Greta Thunberg, the United Nations, and Greenpeace may be among the first associations of climate discourse actors if people were asked on the open street, but that they would be among the first climate posts in the feed is unlikely. By differentiating different centrality measures and examining the dataset, three possible reasons can be identified for this.
Activity: Media visibility can be understood as the product of average visibility per post times the number of posts. It may therefore sound trivial that the number of posts has a major influence. However, the fact that ICCCAD Director Saleemul Huq, with one-hundredth of Thunberg’s five million followers and 30 times as many posts, already has one-third of her visibility may come as a surprise. Of course, he reaches the same users several times with his posts, but assuming each contact and not each person contacted is a starting point of behavioral change and thereby influence, he has gained his prominent role in the network by means of media presence.
Strategic Networking: This is where you can see who uses Twitter’s rules of information dissemination and algorithms most effectively. Like every SN, Twitter rewards the use of its features with increased visibility. Those who make use of indexed hashtags and the various networking options can thus gain attention beyond their followers. An increase in informational power can be achieved in two diametrically different ways. On the one hand, networking with important actors within the relevant filter bubble helps to assume a central role within the cluster. Scientists and environmental organizations in particular make use of this kind of cooperation. Reciprocal networking alliances raise recognition within the targeted audience, which is most likely to disseminates the received information themselves. This form of centrality can best be measured by the Eigenvector centrality. On the other hand, actors can stage themselves as strategic bridge nodes to otherwise unconnected clusters. As the exclusive representatives of the other clusters’ affairs, they enjoy a partial monopoly on information. This type of informational leverage is decoded by high betweenness centrality. Like any other measure, this may be criticized because these bridges could also be mere intermediaries between groups dominated by other actors. Likewise, such a position may testify to exclusion from two neighboring clusters.
Thematic Specialization: As with cooperation and interaction, thematic specialization can also influence placement within the network structure. The IPCC can generate the same centrality with 10% of the posts and 2% of the followers of its institutional parent. The equation would add up if the UN only mentioned climate change with one post in five hundred, but a small sample from the first 100 posts of March 2022 suggests a value in the mid-single digits. This finding is supported by the high centrality of climate activists and environmental organizations.
A large Yellow Cluster dominates the right half of the center. It is home to the authors of the most shared post on climate change and features a collaboration between pop idol Wáng Yībó (王一博) the environmental organization 野生救援WildAid and two Chinese GONGOs. Long after its release in June 2021, the video, which advocates decreasing one’s ecological footprint, experienced a second wave of attention after which it received more than one million reposts.[90] In some places, the cluster has also been able to attract several newspapers and the highly interconnected specialized agencies of the UN. A proximity to the Chinese political and institutional establishment can be assumed, but since many of the nodes have contributed only a single data point, a premature assignment could be wrong and unjust.
To the left is the most homogenous and stable Purple Cluster. It consists of only slightly more than half as many nodes as its yellow neighbor, but it contains 44 of the 200 most central actors discussed later. Instead of sporadic hegemons, there are several interconnected regional powers. Most of the nodes in this cluster represent local ecological ministries (生态环境部，山东环境，临沂生态环境) and media (新华社, 央视新闻，中国日报). The impression from crawling the data is also consistent with this visualization in that the connections are not based on just a few very far-reaching posts but grew continuously over the course of the ten days and resulted from several independent threads. While it can be assumed that other clusters would not have taken a place among the central actors at other times, the steady persistence of media and organizations testifies a particular stability.
As already indicated, the two graphs must be viewed with an urgent caveat. 10 days of data collection is an insufficient basis for drawing generalizations about exactly which actors are influential in the climate debate. As Wáng Yībó demonstrates, actors can achieve considerable reach with just a few posts a year. Between posts, however, they can disappear from the radar altogether. Which climate-affine celebrities would have posted the following 10 days cannot be captured. To derive politically relevant information from the data sets, it is, therefore, necessary to categorize the nodes and rely on stabilization through the law of large numbers. Since there is no reason to suspect that organizations or public figures have different levels of activeness or climate sensitivity at the beginning of March, numbers become more stable, manageable, and comparable.
The following two graphs thus aim to answer research question Q2. They descend from the structural level and provide a more detailed picture of the KOLs of the respective climate debates. To bring some transparency into the categorization process, the account names remain visible. However, the following two graphs are designed to mitigate the high volatility of individual actors. For this purpose, the 200 actors with the highest Eigenvector centrality were identified and entered into a separate data lab.
The resulting graph shows a strong concentration of nodes in the bottom left quarter. This is where most environmental activists and organizations are located. Scientific actors and the media also tend to reside in this area. At the upper end of the graph, public figures begin to appear more frequently. Most of the nodes in the upper third of the graph represent bloggers and public figures. The second most frequent category of this section is constituted by state actors. This surprising phenomenon is only partly a testament to thematic or political proximity. Equally significant is the fact that organizations and environmental activists are less likely to address their content directly to state actors. Political commentators and bloggers, on the other hand, make extensive use of direct responses.
The proximity of the red nodes is particularly striking. On the one hand, this certifies pronounced networking among themselves on the other hand it can be traced back to the overlap of ties to the same organizations and scientists. This shows the advanced stage of development of the climate debate on Twitter. The cluster of activism, science, and institutions has become an unavoidable benchmark. The yellow nodes are found in more or less mixed clouds around this assembly. Even climate deniers cannot avoid references (and thus informational access) to this cluster.
Even the selection of the 200 most central actors cannot prevent that the different sizes of the data sets remains evident. Due to the large number of nodes without any edges at all, the gravity of the Force2 algorithm stretches a symmetric circle. The tendency towards homophily of the individual categories is even more pronounced in this network. Only 11 of the 68 public figures appear in clusters that contain a category other than their own or private accounts. In general, the yellow nodes are the ones with the most edges. Of 137 connections, 64 have them on at least one of their ends. The inward-degree centrality reveals that a large proportion of this entanglement originates from these public figures. Government actors, organizations, and the media, while actively reposted by their neighbors, rarely reciprocate. The rare instances of dissemination in this direction are mostly in the context of direct collaborations.
Organizations and companies are also to be found side by side. Media and state actors are falling out of the role in that they often share a cluster. However, if state media were also decoded as green state actors, this singularity would dissolve.
This table is intended to allow for a quick cross-platform comparison of user categories and the three Booleans. In the following the results are not examined systematically but are reduced to those special features and differentiating characteristics, from which information about political communication can be derived.
a One account suspension between the original data collection and the subsequent extraction of the Verification Boolean.
State Actors: Among the 200 most central nodes, there are more state actors on Twitter than on Weibo. Only after the inclusion of state-operated organizations and state media does it overtake Twitter in terms of governmental nodes. It’s noteworthy that on Twitter, there are only six among the 100 most active, while on Weibo there are 35. This is mainly because Weibo governmental nodes are almost exclusively made up of ministries of the environment, security agencies and state media, while among all categorized state actors there is not even one of institutional nature.
Many politicians are known to have outsourced their digital presence and seek professional support, especially in times of election campaigns. Yet, they do not reach the capacity of institutions. On the other hand, if the number of followers is examined, they suddenly come out very well. Excluding media and GONGOs, Twitter even lists more state actors than Weibo among the 100 most-followed channels. Reasons for that could be the charismatic superiority of individuals over abstract institutions, waves of attention surrounding election events, and varying degrees of competition among the other categories.
Environmental Activism and Science: While those categories can be assigned to several dozen actors under all centrality categories, it is a negligible phenomenon on Weibo. This difference can also be partly explained by a stronger institutionalization of the actors. Scientists are less likely to maintain active accounts themselves but are instead featured by media and organizations. Particularly among the latter, there are four nodes in the top 200 that specialize entirely in aggregating scientific contributions. For activism, however, no comparable absorption by other categories can be observed.
Blogger Dominance: On both platforms, public figures make up the largest category, accounting for a third of the most central actors. That merely shows that the issue of climate change is not imposed by a few active stakeholders but has a prominent role in both net communities. In the case of Twitter, this dominance can also be transferred to the other two metrics. On Weibo, however, the media outperforms them in terms of activity and followers.
Privates: The fact that private accounts have been able to establish themselves better on Weibo can be explained mainly by a Weibo feature that ensures that when posts already reposted by others are shared, an exact copy of the comment is suggested to users. Those who do not personally delete the text and insert a new one thus adopt the integrated @mentionings of the predecessor. Comparatively unknown private accounts can gain some reach as free-rider of a hyped post that way. Those who want to repost on Twitter with a comment will find an empty text field instead. Even if private accounts are equally affected by the absence of uncommented reposts, the increased volatility of small datasets may have played into their hands. For these reasons, it is not possible to draw conclusions on the direction of information flow based on that difference. Though, it can be assumed that the share of the total visibility is much higher than that in the already sorted graphs.
Media: By all measures, traditional news outlets and newspapers are more central players on Weibo than their counterparts on Twitter. However, the major newspapers shine on both platforms when it comes to activity and followers. Of the corresponding top 10, they account for 8 and 7 respectively on Weibo and 8 each on Twitter. China’s well-established newspapers remain highly influential even after going online. However, platforms for aggregating and distributing news and informative videos are also successful. Of the 31 media nodes on Weibo 17 are categorized as governmental.
Climate Denial: Of the two hundred nodes categorized, not even one of Weibo’s Top 200 fell into the category of climate denialism. The cross-checking of the last 30 deleted climate-posts of FreeWeibo also does not contain any denial of man-made climate change. Whether previous waves of censorship and the subsequent anticipatory obedience are responsible for this is difficult to determine empirically. As in the paragraph on activism, Garry King’s theory that discourse cannot get to the point where it is capable of mobilization cannot be contradicted, at least not based on the data. But cultural explanations are also possible. For example, a collective desire for social order, harmony, and the politics of the middle may prevent the Overton window from being overstretched. However, many of Twitter’s 32 denialist actors express a level of opposition that would not be possible on Weibo.
Verification: This data point provides a tiny and unstable starting point for the interrelations of network architecture and network centrality. Active user input through subscriptions, likes, comments, and reposts is losing importance in favor of latent metadata about attention span and digital movement. Platforms, therefore, have their own actorness through the selection of the information they present. Users pay tribute to this agenda-setting power by following the platform’s rules. Obtaining an official verification is just one example of a broader tendency. The extent of the resulting favoritism is difficult to measure, and the high verification rate of both platforms should not be interpreted as regulatory hegemony without further ado. It merely shows that the platforms are more than just blank canvases for showcasing different content. This Boolean can neither distinguish between correlation and causality nor between independent and dependent variables. However, the fact that only 10 of the 31 denialists went through the verification procedure highlights the importance of studying the influence mechanisms of SNs.
Before we move on to answering the research questions, however, three caveats need to be made about the explanatory power of the results.
As soon as the data was scraped, it became clear that neither the 7.1% of people over the age of 13 that Twitter is reaching nor the 9.4% of Weibo are a representative sample of society.[91][92] Users are younger, more urban, more educated, more affluent, more political and, in Twitter’s case, more male than a random selection of their respective countries of origin.[93] The proportion of posts on climate change of some coastal cities in China, for example, is up to five times higher than that of the Chinese population. Beijing, as the home of many media outlets, can claim more than 50% of all climate-related reposts.[94] Even this “small group of people” is difficult to grasp. This is due to the widespread use of multimedia content, which is methodologically almost impossible to capture. Links from images, videos, articles, and websites are omitted in most analyses – including this one – even though 93% of posts on Weibo now contain at least one of them.[95][96] The search terms come with their own set of biases. All results must be read under these signs and locked into the context of the respective SN.
The second caveat concerns the data sets. As mentioned at the beginning, there is no such thing as “raw data”. Each data point arises in a particular context, and two different platforms, used predominantly in different countries and in different languages, offer a variety of potentially confounding variables. We have already seen that seemingly small things, such as suggesting to adopt the comment of the post being reposted, can make a big difference. How big this difference is and how many differences are not even detected remains uncertain. Furthermore, the validity of the comparison suffers from the limited amount of data obtained for the Weibo network. In line with the research gap, more can be said about Twitter and its networks than about Weibo in this analysis.
It should also be noted that every SN is a vibrant melting pot of different actors and interests. On both Twitter and Weibo, individuals, influencers, media, governments, and companies fight their way through the general background noise of the internet in search of entertainment, information, public opinion, and sales markets. The result is a messy emergent system charged with contradictions and intransparencies. For researchers, it is important to accept this complexity and, at the same time, to identify isolated sub-areas within which generalizable hypotheses can be formulated.
Despite these limitations, the network analyses are able to shine some light on the political communication of the two network communities.
Q1: How do the Network Structure of the Climate Debate on Twitter and Weibo compare?
The comparison of political communication first reveals a notable commonality. On both platforms, there is a well-practiced cluster of organizations, media, and state actors in the core establishment. Public figures surround this center from all sides. In the case of Twitter, these peripheral clusters can be differentiated mainly along the political right-left dimension. Within these two clusters, further distinctions can be made. The further actors are squeezed to the edge of the network, the more strongly they stand in opposition to the political status quo. Some highly homogeneous peripheral clusters that are hardly connected to the center show signs of political extremism and climate denial. For the Weibo network, distinguishing between peripheral clusters is more difficult due to the smaller dataset. However, the grouping by the political-apolitical dimension validated by Lin, Hamm, and Reinhardt enjoys considerable explanatory power. Some public figures post more entertaining content about climate change, while others seek political engagement. It is to be expected that this difference would have surfaced more strongly with additional edges. Highly modularized clusters or other indicators of a level of opposition comparable to Twitters cannot be found on Weibo.
The study also identified a number of characteristics that promise activists a particularly central position in the networks. One striking feature is the correlation with activity as measured by the number of posts. On Twitter in particular, some activists and scientists achieve a centrality that is far higher than their number of followers suggests. A second characteristic is strategic networking, which can be understood as effective navigation in the respective network architectures. On both platforms, actors benefit from networking with influential actors of their own or a (thereby) neighboring cluster. A third point can only be attributed to the Twitter network. Here, the thematic specification on climate change seems to give actors visibility beyond the product of their number of followers and the activity. As with Chung and Zhaos paper, actors who are actually only secondarily concerned with climate change attract a great deal of attention on Weibo.
Q2: Which Category of Actors Enjoys the Highest Network Centrality
On both platforms, political commentators, bloggers, and other public figures constitute the largest category among the 200 most central actors. Although the “individual” category already accounted for a large share in Chung and Zhao’s paper, it is even more dominant and, above all, more central in the Weibo elite network. What is completely missing on Weibo, however, are the categories of activism and science. On Twitter, these categories not only make up a quarter of the elite network but are also represented by several nodes with the highest eigenvector and betweenness centrality. This characteristic seems to be a testimony to a general trend of institutionalization on Weibo. The 96 institutional actors of Weibo are contrasted by only 49 on Twitter. This difference is particularly strong when it comes to state actors. Apart from two officials, all 21 state actors on Weibo are state bodies of the CCP. On Twitter, meanwhile, they are all personnel accounts of politicians. While they have high follower counts, they do not match their institutional counterparts in terms of activity. On Twitter, state engagement seems to be less structured and more the product of individual commitment. Beijing, on the other hand, makes clear that it has recognized Weibo as a political arena and integrated its use into its leadership style.
In conclusion, a recommendation for the further complication of the Chinese network community can be made. Scholars of social media must endure just as many theoretical contradictions as the messy networks they encounter. We should not be misled by the strong presence of public figures into one-sided theories of civil liberation, nor should we speak of governmental propaganda in the face of active state institutions. Both simplifications, in their claim to consistency, threaten to impede a deeper understanding of dissemination dynamics. The discipline is simply too young and patchy for an all-encompassing theory of political communication on Weibo. In the course of the research, a number of areas have stood out as particularly pressing.
Methodological frameworks: This research area urgently needs an in-depth analysis of the human-computer interaction of Weibo. The analysis must not only define ways of collecting and evaluating data but also enable meaningful comparisons with other SNs.
Public Availability of Data: To circumvent Weibo’s restrictive data policy, it would be advantageous to collect data not retrospectively, but in real-time. Thus, even uncommented reposts could contribute to network analysis. A data set that is larger and more identical to that of Twitter in terms of its formation would enable a whole series of statistical computations. For example, a multi-factor analysis would be possible, which could be used to make statements about correlations between network centrality and various attributes. Scraping the live feed opens up yet another data point: To what proportions do users see posts from which actors. Long-term scrapes always discriminate against the short-term spikes in attention. Thus, it could be that the identified elite benefits solely from their persistence and is drowned out by a multitude of private accounts at any given time. Since many research projects are hindered by technical bottlenecks, it should be the mission of public institutions to provide such datasets.
Opposition in Chinese Cyberspace: A more detailed analysis of the deleted posts and the periphery of large scale networks would also be interesting. They could provide information about the characteristics of China’s digital opposition. This could be a starting point for the much-Needed Qualitative content analysis around symbolism and terminology of the opposition-conformism dimension.
Governmental Participation: Finally, a deeper analysis of the role of governments in the respective climate debates are needed. Possible research questions could be: With whom do state actors network? Is there a generizable system behind cooperations such as the one between Wáng Yībó and the China Green Carbon Foundation?
In this way, a network of reference values and methodical concepts should be knitted, to make the political communication of the Chinese cyberspace tangible. The reward is a vibrant, colorful, and politically insightful net community of nearly a billion MAUs.
[1] Breuer, Anita / Landman, Todd / Farquhar, Dorothea; 2015; “Social Media and Protest Mobilization: Evidence from the Tunisian Revolution”.
[2] Tandon, Chahat et al.; 2021; “How can we predict the impact of the social media messages on the value of cryptocurrency? Insights from big data analytics”.
[3] Marantz, Andrew; 2020; “Winning: Brad Parscale used social media to sway the 2016 election. He’s poised to do it again”
[4] Kemp, Simon; 2021; “Digital 2021 October Global Statshot Report”; Slide 62.
[5] Gan, Chunmei; 2018; “Gratifications for Using Social Media: A Comparative Analysis of Sina Weibo and WeChat in China”.
[6] Stockmann, Daniela/ Luo, Ting/ Shen, Mingming; 2018; “Designing authoritarian deliberation: how social media platforms influence political talk in China”.
[7] Yang, Yuxin; 2020; “Understanding Young Adults’ TikTok Usage”; p. 43-48.
[8] Fedorenko, Irina/ Sun, Yixian; 2015; “Microblogging-Based Civic Participation on Environment in China: A Case Study of the PM 2.5 Campaign”.
[9] Griffiths, James; 2019; “The Great Firewall of China: How to Build and Control an Alternative Version of the Internet”.
[10] King, Garry; Pan, Jennifer und Roberts, Margaret E.; 2017; “How the Chinese Government Fabricates Social Media Posts for Strategic Distraction, not Engaged Argument”.
[11] The counterpart to follower – list of people who are followed by the given account.
[12] Kemp, Simon; 2021; “Digital 2021 October Global Statshot Report”; Slides 59 and 66.
[13] Walker, Mason/ Matsa, Katerina Eva; 2021; “News Consumption Across Social Media in 2021”.
[14] Lei, Kai et al.; 2018; “Understanding User Behavior in Sina Weibo Online Social Network: A Community Approach”; p. 8.
[15] Daniela Stockmann/ Luo, Ting/ Shen, Mingming; 2019; “Designing Authoritarian Deliberation: How Social Media Platforms Influence Political Talk in China”; p. 3.
[16] Ibid.; p. 4.
[17] Ibid.; p.88
[18] Liu, Lizhi; 2021; “The Rise of Data Politics: Digital China and the World”.
[19] Liang, Fan/ Chen, Yuchen/ Zhao, Fangwei; 2021; “The Platformization of Propaganda: How Xuexi Qiangguo Expands Persuasion and Assesses Citizens in China”.
[20] Sina Weibo Inc.; 2021; “Weibo annual report 2022”; p. 5.
[21] Dembicki, Geoff; 2017; “The Disappearance of Climate Change Denial in China”; p. 63.
[22] Ibid.; 65.
[23] Chung En-Liu; 2015; “Low carbon plot: climate change skepticism with Chinese characteristics”.
[24] Wang, Binbin/ Zhou, Qinnan; 2018; “Climate change in the Chinese mind: An overview of public perceptions at macro and micro levels”.
[25] Liu, Jingfang/ Goodnight, G. Thomas; 2016; “China’s Green Public Culture: network pragmatics and the environment”.
[26] Sánchez Medero, Rubén; 2020; “Democratization in Political Communication”.
[27] Ibid.; p. 617
[28] Jensen, Jens Frederik; 1998; “‘Interactivity’ Tracking a New Concept in Media and Communication Studies”.
[29] Sánchez Medero, Rubén; 2020; “Democratization in Political Communication”; says “the participatory character of this new role depends upon skills and abilities in handling ICT and RRSS”; p. 610.
[30] Johansson, Elena; 2019; “Social media in political communication A substitute for conventional media?”; according to her the new intermediaries strategy is characterized by “central control, a marketing ethos, a master brand, communications cohesiveness, and message simplicity”; p. 151.
[31] Sánchez Medero, Rubén; 2020; “Democratization in Political Communication”; p. 616.
[32] Johansson, Elena; 2019; “Social Media in Political Communication”; p. 1277
[33] Stieglitz, Stefan/ Dang-Xuan, Linh; 2013; “Social Media and Political Communication: A Social Media Analytics Framework”; p. 1286
[34] Sánchez Medero, Rubén; 2020; “Democratization in Political Communication”; p. 609.
[35] Breuer, Anita / Landman, Todd / Farquhar, Dorothea; 2015; “Social Media and Protest Mobilization: Evidence from the Tunisian Revolution”; p. 7.
[36] Zuboff, Shoshana; 2020; “The Age of Surveillance Capitalism – The Fight for a Human Future at the New Frontier of Power”.
[37] Van Dijck, José/ Poell, Thomas; 2013; “Understanding Social Media Logic”, p. 9-11.
[38] Bosetta, Michael; 2018; “The Digital Architectures of Social Media: Comparing Political Campaigning on Facebook, Twitter, Instagram and Snapchat in the 2016 U.S. Elections”; p. 6.
[39] Bradner, Erin; 2011; “Social Affordances: Understanding Technology Mediated Social Networks at Work”.
[40] Daniela Stockmann/ Luo, Ting/ Shen, Mingming; 2019; “Designing Authoritarian Deliberation: How Social Media Platforms Influence Political Talk in China”; p. 5-6.
[41] Bosetta, Michael; 2018; “The Digital Architectures of Social Media: Comparing Political Campaigning on Facebook, Twitter, Instagram and Snapchat in the 2016 U.S. Elections”; p. 9-10.
[42] Jackson, Linda A./ Wang, Jin-Liang; 2013; “Cultural differences in social networking site use: A comparative study of China and the United States” compared how students in the United States and China used social media, to find that Americans prioritized quantity over closeness of contacts in contrast with their Chinese counterparts, suggesting differences in users’ approaches to sociability within social media.
[43] Kemp, Simon; 2021; “Digital 2021 October Global Statshot Report”; Slide 62.
[44] Sayce, David; 2020; “The Number of Tweets per Day”.
[45] Twitter Inc.; 2021; “Fiscal Year 2020 – Annual Report”; p. 6.
[46] Peddinti, Sai Teja/ Ross, Keith W./ Cappos, Justin; 2017; “User Anonymity on Twitter”; p. 84-85
[47] Ibid.; p. 86
[48] Bosetta, Michael; 2018; “The Digital Architectures of Social Media: Comparing Political Campaigning on Facebook, Twitter, Instagram and Snapchat in the 2016 U.S. Elections”; p. 9.
[49] Ibid.; p. 8.
[50] Johansson, Elena; 2019; “Social Media in Political Communication”; p. 152.
[51] Bosetta, Michael; 2018; “The Digital Architectures of Social Media: Comparing Political Campaigning on Facebook, Twitter, Instagram and Snapchat in the 2016 U.S. Elections”; p. 28.
[52] Ibid.; p. 10.
[53] Koumchatzky, Nicolas/ Andryeyev, Anton; 2017; “Using Deep Learning at Scale in Twitter’s Timeline”.
[54] Bosetta, Michael; 2018; “The Digital Architectures of Social Media: Comparing Political Campaigning on Facebook, Twitter, Instagram and Snapchat in the 2016 U.S. Elections”; p. 27.
[55] Kemp, Simon; 2021; “Digital 2021 October Global Statshot Report”; Slide 62.
[56] Crouch, Erik; 2016; “Here’s what happens every minute online in China”; It can be assumed that the number last issued by Weibo in 2015 has increased substantially since then.
[57] Sun, Huatong; 2013; “Sina Weibo of China: From a Copycat to a Local Uptake of a Global Technology Assemblage”; p. 27.
[58] Ibid.; p. 32.
[59] Wang, Qingning; 2021; “The Chinese Internet – The Online Public Sphere, Power Relations and Political Communication”; p. 107.
[60] Wan, Vanessa; 2018; “The Ultimate Guide to Sina Weibo: The Largest Micro-Blogging Platform in China”.
[61] Weibo Official; 2018; in: “Explanation of Weibo’s Traffic Selection Mechanism”; [transl. 微博限流机制解析].
[62] Sina Weibo Inc.; 2021; “Weibo annual report 2022”; p. 15.
[63] Yang, Yi et al.; 2021; “Cross-platform comparison of framed topics in Twitter and Weibo: machine learning approaches to social media text mining”; p.73-76.
[64] Newman, Todd P.; 2017; “Tracking the release of IPCC AR5 on Twitter: Users, comments, and sources following the release of the Working Group I Summary for Policymakers”.
[65] Lin, Zihao/ Hamm, Andrea/ Reinhardt Susanne; 2018; “Political Communication Chinese Style: The Elite Network in State-Regulated Sina Weibo”; p. 92.
[66] Ibid,; p. 94.
[67] Ibid,; p. 96.
[68] Ibid,; p. 107.
[69] Liu, John Chung-En/ Zhao, Bo; 2017; “Who Speaks for Climate Change in China? Evidence from Weibo”; p. 416.
[70] Ibid.; p. 421.
[71] Ibid.; p. 421.
[72] Jacomy, Mathieu et al.; 2014; “ForceAtlas2, a Continuous Graph Layout Algorithm for Handy Network Visualization Designed for the Gephi Software”; p.2.
[73] Lin, Zihao/ Hamm, Andrea/ Reinhardt Susanne; 2018; “Political Communication Chinese Style: The Elite Network in State-Regulated Sina Weibo”.
[74] Ibid.; p.86-93.
[75] Stieglitz, Stefan/ Dang-Xuan, Linh; 2013; “Social Media and Political Communication: A Social Media Analytics Framework”; p. 1283.
[76] Cambridge Intelligence; 2020; In: “Social Network Analysis and Visualization”; p. 6-8.
[77] Ibid.; p. 8.
[78] Lin, Zihao/ Hamm, Andrea/ Reinhardt Susanne; 2018; “Political Communication Chinese Style: The Elite Network in State-Regulated Sina Weibo”; p. 92.
[79] Secondary ranking based on participation in Weibo-features such as the microevents, trends, and overall usage time.
[80] Twitter Description, which can be found in the dataset forwarded at the end of this paper
[81] If a graphic with these actors was to be created, it is best to simply use the in-degree centrality or the follower count.
[82] Wang, Yibo [王一博]/ 野生救援WildAid/ China Green Carbon Foundation; 2021; Video about plastic waste in the oceans and ways to reduce usage under the hashtag “Vitality comes from action”; [transl. 生动源自行动].
[83] Due to the use of different centrality metrics, a total of 369 accounts were categorized. 32 of them are state actors.
[84] Kemp, Simon; 2021; “Digital 2021 October Global Statshot Report”; Slide 119.
[85] Ibid.; Slide 132.
[86] Johansson, Elena; 2019; “Social Media in Political Communication”; p. 157.
[87] Liu, John Chung-En/ Zhao, Bo; 2017; “Who Speaks for Climate Change in China? Evidence from Weibo”; p. 416.
[88] Liu, Jun/ Zhao, Jingyi; 2021; “More than plain text: Censorship deletion in the Chinese social media”.
[89] Sina Weibo Inc.; 2018; “2018 Sina Media White Paper”; [transl. 2018 新浪媒体白皮书]; p. 423.
[90] Wang, Yibo [王一博]/ 野生救援WildAid/ China Green Carbon Foundation; 2021; Video about plastic waste in the oceans and ways to reduce usage under the hashtag “Vitality comes from action”; [transl. 生动源自行动].
[91] Kemp, Simon; 2021; “Digital 2021 October Global Statshot Report”; Slide 119.
[92] Ibid.; Slide 132.
[93] Johansson, Elena; 2019; “Social Media in Political Communication”; p. 157.
[94] Liu, John Chung-En/ Zhao, Bo; 2017; “Who Speaks for Climate Change in China? Evidence from Weibo”; p. 416.
[95] Liu, Jun/ Zhao, Jingyi; 2021; “More than plain text: Censorship deletion in the Chinese social media”.
[96] Sina Weibo Inc.; 2018; “2018 Sina Media White Paper”; [transl. 2018 新浪媒体白皮书]; p. 423.
© all rights reserved | 2024
