AbstractDespite growing recognition of social bots’ role in swaying public opinion, evidence of their specific impact on public cognition remains limited. Applying the network agenda-setting theory and multiple regression quadratic assignment procedure (MRQAP) analysis of X (formerly Twitter) posts, we examine the relative impact of social bots and legacy media on public agenda networks over time, focusing on China’s dual-carbon policy. Findings reveal that social bots’ agenda networks exert a significant influence on public agenda networks, with their impact being more pronounced than that of legacy media during the initial period (T1). However, this influence diminishes in the subsequent period (T2), suggesting that social bots have a potent yet short-lived effect on shaping public cognitive networks around specific issues. Furthermore, social bots exhibited more negative attitudes towards dual-carbon topics, contrasting with the neutral stance of legacy media. A comparative semantic network analysis highlights the distinct narrative structures promoted by social bots and legacy media. This research provides crucial insights into automated actors’ roles in molding environmental discourse and public perception.
IntroductionThe rapid digitalization of the media landscape has given rise to a complex, hybrid ecosystem where legacy news outlets operate alongside emergent online communicators, fundamentally altering the dynamics of information flow and public opinion formation (Chadwick, 2017). Central to understanding these transformations is the theory of agenda-setting, which posits that media wield significant influence over public attention and perception by strategically curating the salience of issues and their attributes (McCombs, 2005; McCombs and Shaw, 1972). As the constellation of actors shaping public discourse expands to include not only human journalists but also automated software agents known as social bots (Ferrara et al., 2016; Luo et al., 2023; Zhang et al., 2024), agenda-setting research faces new imperatives to interrogate the evolving mechanisms of media influence in the digital age.The network agenda-setting (NAS) model has emerged as a crucial theoretical lens for investigating the complex agenda-building dynamics of contemporary media environments (Guo and McCombs, 2011; Guo, 2013). Drawing upon the associative network model of memory in cognitive psychology (Anderson, 1983), NAS conceptualizes media effects as a process of constructing and transferring the salience of interconnected semantic networks that define how audiences comprehend and prioritize relationships between issues, attributes, and actors. By representing the multidimensional structures of media and public agendas as complex networks of concepts, the NAS framework allows researchers to systematically trace the dissemination of associative relevance between diverse communication sources and their audiences with exceptional granularity (Guo and Vargo, 2015, 2017). Numerous studies have employed the NAS model to investigate agenda-building processes across various domains, including political contexts (Vargo et al., 2014), public health crises (Tahamtan et al., 2022), and the impact of emerging media technologies (Guo and Liu, 2022). For example, Guo and Vargo (2015) demonstrated that traditional news media exerted a stronger influence on public perceptions of issue networks than political campaign messages, highlighting the media’s ability to transfer both issue networks and affective associations to public cognition. Vargo et al. (2014) showed how different types of news media shaped the network agendas of candidate supporters on Twitter. Similarly, in the Korean context, Kim et al. (2024) found that conservative YouTube channels had a weaker impact on public issue networks compared to traditional news media and liberal channels. These examples demonstrate the wide-ranging applications of the NAS model in uncovering the complex agenda-building dynamics that characterize contemporary media environments.Traditionally, legacy media outlets, such as newspapers, television networks, and radio stations that have long-standing reputations in journalism, have been regarded as primary agenda setters (McCombs and Shaw, 1972). However, the meteoric rise of social bots - sophisticated algorithms designed to emulate human users and autonomously generate high-volume content streams aimed at steering online discourse (Ferrara et al., 2016)—signifies a formidable new player in the agenda-setting matrix. A growing body of literature attests to the capacity of social bots to propel information virality, amplify misinformation and disinformation, and fuel polarization around contentious sociopolitical issues (Aldayel and Magdy, 2022; Yuan et al., 2019). However, empirical studies systematically assessing the direct agenda-setting efficacy of bots versus legacy media remain notably sparse (see Zhang et al., 2024), leaving critical questions unanswered about the potential of automated accounts to manipulate public salience networks, especially in comparison to the established clout of elite journalistic institutions.To address this gap, our study employs computational techniques, including semantic network analysis and Multiple Regression Quadratic Assignment Procedure (MRQAP), within the NAS framework to conduct a comparative examination of the network agenda-building capacities exhibited by social bots and legacy media within the consequential context of China’s dual-carbon policy discourse. We aim to: (a) quantify the relative influence of bot-driven versus media-driven agenda-setting effects on shaping the salience network of public discussions and (b) unravel the semantic and affective dimensions of their respective narrative strategies to pinpoint potential divergences in how automated and journalistic sources construct frames around the focal issue. Through this approach, the study provides novel empirical insights illuminating the evolving interplay between human, media and AI agenda setters in today’s hybrid media environment.Literature reviewThird-level agenda setting theory: the network agenda setting modelAgenda-setting theory, which postulates that the media can influence the salience of issues and attributes in the public’s mind, has evolved substantially since its inception over half a century ago. The pioneering Chapel Hill study by McCombs and Shaw (1972) empirically verified Walter Lippmann’s (1922) seminal notion of the media constructing a “pseudo-environment” that shapes “the pictures in our heads.” This foundational idea catalyzed agenda-setting’s emergence as one of the most significant theories in communication effects research (McCombs, 2005). The initial conception, termed the first level of agenda-setting, focused on how the media transfer the salience of issues or objects to influence the public agenda (Vargo, 2018). Subsequent research uncovered a second level—beyond merely priming issue salience, the media also emphasizes specific attributes of those objects, transferring salience to particular attribute agendas in public cognition (Guo et al., 2012).However, cognitive psychology posits that human understanding of social reality occurs not in a linear fashion, but through interconnected network-like structures (Guo and McCombs, 2011). Recognizing this limitation in the first two levels, Guo and McCombs (2011) proposed the third-level agenda-setting theory: the NAS model. This perspective theorizes that the media does not merely convey objects or attributes, but can also transfer salience to the associative connections between attributes, shaping the cognitive networks underlying how the public conceptualizes issues.The NAS model provides a framework for investigating how media influences construct and propagate semantic networks of associated attributes in the minds of audiences. Its core proposition moves agenda-setting research closer to capturing the rich “pictures” metaphorically described by Lippmann (Guo and McCombs, 2011). Empirical studies have validated the model’s assumptions, demonstrating how media can bundle objects and attributes into associative clusters, shaping how audiences think about issues (Guo and McCombs, 2011; Guo, 2013). For example, significant correlations were found between newspaper coverage, which attributed different personal qualities to two political candidates (media network agenda), and the public’s perception of these candidates (public network agenda) in both the 2002 and 2010 Texas gubernatorial elections (Guo and McCombs, 2011; Guo, 2013).As digital technologies reshape the media landscape into “hybrid media systems” (Chadwick, 2017), the theoretical prominence of the NAS model has grown. The NAS model has demonstrated applicability beyond traditional media, extending to analyses of new media dynamics. Researchers leveraging the NAS model have investigated agenda network correlations across various media forms. For instance, Kim et al. (2024) examined the relationship between issue network structures provided by news media and political YouTube channels, as well as the influence of these structures on the public’s perceived issue networks. Using network visualizations, Su and Hu (2020) identified a reciprocal yet asymmetrical relationship between newspapers and Twitter in the context of the Diaoyu/Senkaku Islands dispute.Furthermore, social media has diminished legacy gatekeeping constraints while broadening the range of actors who can shape political discourse and public agenda networks (Posegga and Jungherr, 2019), thereby introducing additional complexity to agenda-setting dynamics (Gilardi et al., 2022). Researchers have also applied the NAS model to examine the roles of different agenda setters within the same digital platform. Vargo et al., (2014), for example, demonstrated NAS effects between diverse news media types and candidate supporters using computational methods and extensive Twitter datasets. Similarly, Guo and Liu (2022) explored the NAS effects of hashtag activism, offering new insights into the NAS model.This increasingly complex and decentralized media ecology underscores the growing importance of theoretical frameworks like the NAS model, particularly as new communication actors emerge. Algorithmic intelligent agents are exerting an increasing influence on public opinion, often in ways that traditional agenda-setting research has yet to fully address (Zhao et al., 2024). Despite the advancements in NAS research highlighted earlier, there has been no systematic investigation into these new communicators as agenda setters. Social bots, as autonomous algorithmic agents, are fundamentally transforming social networks and the communication landscape. Understanding how these bots construct agenda networks and influence media or public opinion networks represents a critical gap in contemporary agenda-setting research.Social bots as emerging agenda settersThe rise of artificial intelligence has introduced social bots as a new class of communicators coexisting with humans in online ecosystems (Zhao et al., 2024). Social bots are automated accounts that are controlled by computer algorithms (Ferrara et al., 2016; Woolley and Howard, 2016). They engage in discussions on topics such as politics, economics, health, and emergencies by automatically generating content, imitating human behavior, and interacting with humans (Zhang et al., 2024; Broniatowski et al., 2018; Yuan et al., 2019). Visually, social bot accounts are indistinguishable from human accounts, as they also include basic information such as account ID, username, avatar, and personal signature. Some accounts even display the registered geographic location. Through interactions like liking, following, and forwarding, they can form social networks with other users (Zhao et al., 2024). Ordinary users cannot easily identify these automated accounts. Their distinguishing characteristics - large dissemination capacity (Ferrara et al., 2016), full autonomy (Woolley and Howard, 2016), and covert operational potential (Rheault and Musulan, 2021) - have positioned social bots as powerful instruments for shaping public opinion. Empirical studies have documented their significant presence: during the 2016 U.S. presidential election, approximately 400,000 social bots generated about 3.8 million tweets, constituting roughly 20% of election-related discussion (Bessi and Ferrara, 2016). Prior research has documented social bots’ ability to amplify information spread and manufacture artificial popularity around particular issues. During major political events, bots have demonstrated significantly higher viral spread rates than human accounts and can rapidly reach large audiences within a short time frame (Fahmy et al., 2023). Social bots have also been widely deployed to disseminate misinformation, with studies showing they substantially increase the spread of false narratives compared to human-only networks (Shao et al., 2018). Moreover, bots can exacerbate polarization by propagating emotionally-charged rhetoric, biased viewpoints, and inflammatory speech around social issues like the COVID-19 pandemic (Himelein-Wachowiak et al., 2021; Shi et al., 2020) as well as contentious topics like vaccine advocacy (Zhang et al., 2022).While these studies corroborate social bots’ impacts on public discourse from various angles, systematic empirical evidence examining whether and how they directly influence general public opinion remains lacking. In other words, there are open questions around whether these non-human actors are capable of commanding agenda salience and attribute prioritization. Existing research suggests social bots may be effective at setting agendas due to characteristics like higher message frequencies, wider dissemination, and targeted following/networking behaviors (Ross et al., 2019; Zhang et al., 2024). Zhang et al. (2024) found that bots can successfully propagate narratives through high-frequency messaging, but did not measure whether this actually shifted public cognitive network. Conversely, lower credibility perceptions of bot accounts could diminish their agenda-setting influence compared to mainstream sources (Shao et al., 2018). Additionally, no studies have directly compared the agenda-setting effectiveness of social bots versus legacy media, leaving questions about their relative influence unanswered.Recent scholarly efforts have begun exploring social bots’ potential agenda-setting dynamics, comparing their agenda networks to media and human counterparts in specific contexts like elections (Zhang et al., 2024) and geopolitical conflicts (Zhao et al., 2024). However, these works have primarily adopted linear temporal approaches analyzing whose agendas precede others, lacking investigations into the deeper mechanisms through which bots may mold public cognitive networks.In today’s hybrid media ecosystem where humans and artificial agents collaboratively shape online discourse, it is imperative to systematically examine “the pictures in our heads” in relation to the “pseudo-environment” that includes social bots. Integrating the NAS model with advanced computational methods like the MRQAP offers a framework for investigating whether and how social bots exert agenda-setting influences distinct from legacy media outlets. Our study bridges these gaps by examining both the relative influence of bot-driven versus media-driven agenda-setting effects and their evolution over time, providing empirical evidence in the context of environmental policy discourse. This study therefore proposes the following research question:RQ1: Do public issue networks on China’s dual-carbon policy align more closely with social bots’ issue networks or legacy media’s issue networks?Issue networks of social bots and legacy media on dual-carbon issuesGlobal climate change is one of the most pressing environmental challenges of the 21st century, and carbon neutrality has become a critical component of global climate governance (Li et al., 2023). Carbon neutrality refers to balancing anthropogenic carbon dioxide emissions with equivalent removals over a specified period. In line with this concept, the notion of “carbon peak” has also been introduced, defined as the point at which carbon dioxide emissions reach their highest level, stabilize, and subsequently begin to decline. Achieving carbon peak is a prerequisite for attaining carbon neutrality. In September 2020, as a rapidly developing nation, China announced its “dual-carbon” policy, committing to reach carbon peak by 2030 and achieve carbon neutrality by 2060. This policy represents a significant strategic initiative to address climate change and fulfill international commitments (Wang et al., 2021).This policy has attracted widespread attention both domestically and internationally, due to its significant implications for global climate governance, economic development, and geopolitical dynamics (Kong and Wang, 2022). From a global viewpoint, the “dual-carbon” goal extends beyond a mere climate change governance concept, embodying a critical issue with multiple dimensions, including science, politics, and diplomacy. It also serves as a “hidden front” laden with conflicts, confrontations, and strategic interactions among various entities vying for international discourse power (Mu et al., 2022). Thus, the complex nature of the dual-carbon policy presents an ideal case study for examining the interplay between legacy media and social bots in shaping public discourse.Social media platforms have emerged as critical spaces for climate change discourse (Gokcimen and Das, 2024; Pearce et al., 2019), providing opportunities for diverse stakeholders to engage in discussions (Lin and Kim, 2023; Mooseder et al., 2023). For example, Yao et al. (2022) conducted social network and sentiment analyses of English and French tweets about carbon neutrality on Twitter. Their findings highlight significant public engagement with the dual-carbon issue, though both French-speaking and English-speaking users predominantly express negative sentiments about it. In contrast, research on Chinese social media reveals distinct patterns. Zeng et al. (2022) identified that Weibo users prioritize green development and construction industry practices. Similarly, Li et al. (2023) observed predominantly positive sentiment (83.9%) among Chinese users discussing carbon neutrality on Weibo. Moreover, information shared on social media plays a pivotal role in shaping public understanding of climate change (Singh et al., 2020). While some studies suggest that media promotion can influence perceptions and behaviors related to carbon neutrality (Lu and Wang, 2018), others emphasize the greater impact of non-media actors engaging the public through online platforms in raising awareness (Arlt et al., 2023).Amidst this complex landscape, social bots have emerged as influential actors in climate change discourse. During the U.S. withdrawal from the Paris Agreement, social bots amplified support for the decision and propagated climate change denial narratives on Twitter (Marlow et al., 2021). Chen et al. (2021) found that social bots actively participate in climate change discussions, contributing a significant portion of the content and exhibiting strategic behavior in triggering debates and mobilizing like-minded users. Given the pseudonymous nature of bot operators and the absence of editorial standards, social bots have the potential to propagate more extreme emotional sentiments and skewed narratives compared to legacy media, which are generally expected to adhere to professional norms of objectivity and impartiality (Ojala, 2021). In the context of China’s dual-carbon policy, it is plausible that social bots may construct issue narratives with clearer persuasive intents, propaganda frames, and distinctive linguistic signatures. For instance, bot networks with a vested interest in promoting or discrediting the policy may employ highly emotive language, cherry-picked evidence, or conspiracy theories to sway public opinion (Shao et al., 2018; Zhang et al., 2022). Conversely, legacy media outlets, guided by journalistic principles, are more likely to present a balanced coverage of the policy, incorporating diverse perspectives and relying on authoritative sources (Benham, 2020).Building on earlier studies (e.g., Yao et al., 2022) that analyzed carbon neutrality discussions on platforms like Twitter, our study advances this research in three significant ways. First, we focus on the role of social bots as emerging agenda setters in shaping public perceptions of China’s dual-carbon policy, moving beyond traditional human-centered approaches. Second, we employ a novel methodological combination of the NAS framework and MRQAP analysis to quantitatively assess the relative influence of different information sources (social bots and legacy media) on public cognitive networks. Third, we examine the temporal dynamics and emotional narrative framing of bot influence, revealing how their agenda-setting power evolves over time in comparison to traditional media. These insights illuminate the sustainability of bot-driven narratives relative to legacy media coverage.To contextualize these contributions, we propose the Hybrid Agenda-Setting Dynamics Model as a conceptual framework (see Fig. 1). This model highlights the distinct mechanisms through which legacy media and social bots influence public understanding of China’s dual-carbon policy. Legacy media exerts influence by adhering to professional norms, emphasizing balanced narratives on policy implementation, economic impacts, and international collaboration. Social bots, on the other hand, shape public discourse through high-frequency messaging and emotional amplification, often promoting polarized narratives designed to sway opinion. Applying this framework to China’s dual-carbon policy reveals systematic differences in the issue networks constructed by these two actors. Legacy media tends to build balanced and credible narratives, while social bots amplify emotive and polarized frames that may challenge institutional perspectives. Given these distinctions, our research focuses on how these communicators differ in shaping emotional attitudes and narrative structures. This leads to the following research question:Fig. 1Conceptual model of MRQAP.Full size imageRQ2: How do social bots’ and legacy media’s issue networks differ in their (a) emotional attitudes and (b) narrative structures when discussing China’s dual-carbon policy?MethodsData collection and sampleThis study focused on tweets related to China’s “dual-carbon” goals on Twitter. Twitter’s open API and historical encouragement of third-party applications have led to a significant presence of social bots on the platform (Gorwa and Guilbeault, 2020). Studies have shown that bots produce a substantial portion of tweets on various topics, including political and climate issues (Marlow et al., 2021; Zhang et al., 2024). This prevalence, combined with Twitter’s role in shaping public opinion on global issues like climate change, makes it an ideal environment for studying how automated accounts influence public discourse. To gather the relevant tweet data, key hashtags and search queries were determined based on observed Twitter discussions. The hashtags included #China, #Carbonpeak, #Carbonneutral, #Climatechange, and #Globalwarming. Consequently, four search strings were formed: “#China (AND) #Carbonpeak”, “#China (AND) #Carbonneutral”, “#China (AND) #Climatechange”, and “#China (AND) #Globalwarming”. The data collection period spanned 1 month, from November 1 to November 30, 2022. This timeframe coincided with the 27th United Nations Climate Change conference (COP 27), a period of heightened global focus and discussion on climate change, making it a representative moment for concentrating on China’s climate change management and behaviors.Data was collected using Netlytic, a cloud-based platform for text data mining and social network analysis, widely utilized in social science research (Santarossa et al., 2019). Following Netlytic’s guidelines, the specified search strings were inputted to collate relevant tweets via a specialized Twitter API. The raw data underwent cleaning processes, removing duplicate tweets and irrelevant information, retaining essential elements such as text, author, time of posting, user characteristics, and relational data like retweets and replies. The final dataset consisted of 37,119 valid tweets from 30,169 Twitter users.Social bot detectionThis study employed Botometer, a popular online detection tool developed by Indiana University’s research institution. Botometer, using a supervised learning algorithm and a random forest as its underlying logic, examines over 1150 specific features including user characteristics, network features, temporal patterns, tweet characteristics, and emotional features to train and model (Davis et al., 2016).Footnote 1 The tool provides a score (bot score) between 0 and 5, with higher scores indicating a greater likelihood of an account being a bot. Using Botometer’s official API and Python code, the study conducted batch detection on the 30,169 Twitter users in the dataset, obtaining scores for 23,170 accounts.Footnote 2Drawing on previous studies (Yang et al., 2019; Zhao et al., 2024), accounts scoring 3 or above were classified as social bots. This resulted in 6,075 accounts being identified as social bots, constituting about 26.2% of the detectable accounts. Following the detection of social bots within the dataset, we proceeded to examine the remaining Twitter accounts to distinguish between ordinary users and legacy media organization accounts. We operationalized legacy media as established news organizations with a history of traditional print or broadcast journalism that now also publish content online. Specifically, we included: (a) National and international newspapers with significant digital presence (e.g., The New York Times, The Guardian, People’s Daily), (b) Major television news networks (e.g., CNN, BBC, CCTV) and (c) Established news agencies (e.g., Reuters, Associated Press, Xinhua). Three researchers were assigned to identify media organization accounts based on account descriptions and verification badges on their Twitter profiles. These accounts were also tagged with the respective countries of the media organizations. Through this process, we were able to identify and confirm a subset of tweets that included 329 tweets from media organizations, 6,075 tweets from social bots, and 16,766 tweets from human users. In addition, for lagged regression analysis, we collected an additional 3825 tweets from human users from December (Time 2) constituting a lag-time public issue network.AnalysisMultiple regression quadratic assignment procedure (MRQAP)To address Research Question 1 (RQ1), our study employed the MRQAP analysis method, which has been previously utilized in NAS research (Guo and Vargo, 2015; Kim et al., 2024). We conducted the analysis using Python’s “MRQAP” module (Kim et al., 2024), which facilitated efficient computation of large matrices and numerous permutations.Following the approach of Kim et al. (2024), we extracted and analyzed 1,610 common words and frequencies from tweets of social bots, legacy media, and the public. Utilizing the top 100 of these common core words as nodes, we constructed the issue network matrices. The edges in these matrices represented the connections between each core word, forming a network of associations based on their co-occurrence in the tweets. These matrices were designed to represent the respective agenda networks of each group, offering a structural depiction of how elements within these networks were associated. The public’ issue network matrix served as the dependent matrix, while those of social bots and legacy media functioned as independent matrices. We implemented 10,000 permutations for the MRQAP analysis, randomly shuffling rows and columns of the dependent variable matrix while maintaining the structure of independent matrices. This technique allows for simultaneous comparison of multiple network matrices (Dekker et al., 2007), enabling quantification of bot-driven and media-driven influences on public perceptions while controlling for confounding factors.Sentiment analysis measurementTo compare the emotional responses of social bots and legacy media to the topic of China’s dual-carbon goals, we utilized the XLM-T natural language processing pre-trained model developed by Barbieri et al. (2021). The XLM-T model is a large language model based on the transformer architecture, pre-trained on 200 million multilingual tweets and specifically tailored for sentiment classification tasks. This model is available on the Hugging Face platform (https://huggingface.co). In our study, we accessed this model via API to perform sentiment analysis on the tweet data from both legacy media and social bots.Machine coding was used to generate sentiment labels (i.e., positive, neutral, and negative) and the probability scores for each sentiment category (ranging from 0 to 1, with higher scores indicating a greater likelihood of belonging to that sentiment category). To ensure the reliability and validity of the machine coding results, a random sample of 1000 comments was manually annotated. Subsequently, the consistency between machine and human coding was assessed using the Holsti coefficient (Holsti, 1969). The intercoder consistency for sentiment (Holsti coefficient = 0.863) demonstrated satisfactory results, thereby ensuring the accuracy and reliability of machine coding.Semantic network analysisTo address Research Question 2 (RQ2), our study employed semantic network analysis to compare the narrative differences between social bots and legacy media in their discussion of China’s dual-carbon goals. Semantic network analysis is a method used to explore the relationships between concepts within a text, providing a structural view of how ideas and themes interconnect. This approach is particularly useful for uncovering the underlying semantic structures and narrative frames in large volumes of text data.For our semantic network analysis, we use python to process and analyze the text from the tweets. This involved extracting key terms and phrases (Top 100), identifying their frequencies, and determining the connections between them based on their co-occurrence within the same tweet. The process resulted in a network of terms, where nodes represented the terms and edges signified the connections between them. We separately analyzed the networks for tweets from social bots and legacy media to draw comparisons. We used the Louvain module to form clustering networks based on betweenness centrality (Hallinan et al., 2021; Segev, 2020). Finally, we used network analysis software (Visone) to visualize the networks.ResultsEffects of legacy media and social bots on public issue networksOur first research question (RQ1) focused on how the issue networks of legacy media and social bots influence the public’s issue networks. To address this, we conducted an MRQAP analysis, which revealed a dynamic shift in influence over time. MRQAP results (see Table 1) showed that in the initial period (Time 1, November), social bots exhibited a remarkably strong positive impact on the public agenda, with a coefficient of 0.87 (SE = 0.01, p < 0.001, t = 164.20). This coefficient indicates that for every unit increase in the salience of an issue in the social bot network, there was a corresponding 0.87 unit increase in the public agenda network, suggesting a powerful initial agenda-setting effect of social bots. In contrast, legacy media showed a more modest, though still significant, positive impact during this initial period, with a coefficient of 0.09 (SE = 0.01, p < 0.001, t = 18.00). This indicates that while legacy media did influence the public agenda, its impact was considerably smaller than that of social bots in the early stages of discourse formation.Table 1 MRQAP on Public Issue Network.Full size tableHowever, this dynamic underwent a dramatic shift in the subsequent period (Time 2, December). The influence of social bots not only weakened but reversed direction, becoming slightly negative with a coefficient of −0.06 (SE = 0.01, p < 0.001, t = −5.56). This suggests that in the later period, increases in issue salience in the social bot network were associated with small decreases in salience in the public agenda network, indicating a potential backlash or fatigue effect. Conversely, the influence of legacy media grew substantially stronger in this later period, with the coefficient increasing to 0.81 (SE = 0.01, p < 0.001, t = 65.13). This marked increase suggests that as time progressed, the public agenda became much more aligned with the narratives presented by legacy media outlets. Figure 2 shows the results of the distribution of 10,000 random permutations, showing the robustness of the regression results.Fig. 2: Distribution of 10,000 randomized permutations.Note: The permutation test distributions from the MRQAP analysis provide a randomized baseline against which to compare the observed regression coefficients for social bots and media outlets. These distributions were generated from 10,000 permutations, representing the potential range of coefficients if the associations were due to chance alone.Full size imageSentiment analysis of legacy media and social botsOur second research question (RQ2) examined the differences between legacy media and social bots in their emotional attitudes and narrative networks regarding China’s dual-carbon policy. To answer this, we conducted sentiment analysis and semantic network analysis.The sentiment analysis uncovered stark differences in the emotional tone of content disseminated by social bots and legacy media (see Fig. 3). Social bots demonstrated a strong tendency towards negative sentiment, with an average probability for negative expressions of 0.54 (SD = 0.34). This is significantly higher than their probabilities for positive (M = 0.19, SD = 0.23) or neutral (M = 0.27, SD = 0.19) sentiments. The dominance of negative sentiment in bot-generated content suggests a potential strategy of polarization or amplification of critical narratives surrounding China’s environmental policies.Fig. 3Sentimental differences between social bots and legacy media.Full size imageLegacy media, on the other hand, exhibited a more balanced sentiment distribution. The probabilities for neutral (M = 0.36, SD = 0.18) and negative sentiments (M = 0.36, SD = 0.31) were nearly identical, with positive sentiments only slightly less prevalent (M = 0.28, SD = 0.24). This equilibrium in sentiment suggests a more measured approach by legacy media in discussing China’s environmental initiatives.Exploring issue networks of legacy media and social botsOur semantic network analysis revealed significant structural differences between the issue networks of legacy media and social bots. Legacy media networks (see Fig. 4) were characterized by six distinct clusters, with the main cluster centered around the COP27 conference. This cluster included diverse topics such as “green”, “energy”, “meeting”, “effort”, “deal”, and “development”, indicating a multifaceted approach to climate discourse. Additional clusters in the legacy media network highlighted themes of international cooperation, economic development, and global leadership, suggesting a narrative that emphasizes collaborative approaches to climate challenges.Fig. 4Associative issue network of news organizations.Full size imageIn contrast, social bots’ issue networks (see Fig. 5) formed three main clusters with a notably denser structure. The primary cluster in the bot network, while including general terms like “climate” and “change”, also prominently featured “Biden” and “China”. Surrounding the “China” node, we observed terms like “pay”, “polluter”, “reparations”, and “largest”, indicating a more accusatory narrative that frames China as a major polluter responsible for climate damage. For example, a tweet with a high negative sentiment score in the social bot corpus states, “on the climate change reparations nonsense, I find it galling that China, who are doing far worse things in the developing world than Britain ever did, centuries ago, are not only the worst current polluters but seem to get be getting off scot free.” This tweet was replicated by 164 social bot accounts. Moreover, social bots portray a negative view of the COP27 meeting. The second cluster, containing “COP27”, “loss”, “damage”, “worst”, and “agenda”, suggests criticism of the conference outcomes or emphasizes the severity of environmental damage. Terms like “worst” and “loss” likely underline the negative aspects of environmental degradation and disapproval of the conference’s efforts.Fig. 5Associative issue network of social bots.Full size imageOverall, legacy media’s diverse network presents a balanced narrative on environmental issues, spanning from human and economic impacts to international diplomacy and policy. In contrast, social bots’ networks adopt an accusatory tone, focusing on geopolitical and environmental responsibilities and highlighting China’s perceived duties, thus promoting a blame-centric narrative instead of collaborative problem-solving.Table 2 shows the differences in network attributes between legacy media and social bots. Social bot networks displayed a higher average degree (53.08 vs. 37.58 for legacy media), indicating denser conceptual interconnections and potentially more efficient pathways for information flow. The shorter average path length in social bot networks (1.46 vs. 1.62) suggests tighter semantic proximity, facilitating quicker associative activation between concepts. Moreover, the higher network density (0.53 vs. 0.38) and average clustering coefficient (0.74 vs. 0.659) in social bot networks indicate tightly clustered information ecosystems, which may facilitate echo chamber effects by amplifying specific narratives or viewpoints within highly interconnected semantic contexts.Table 2 Differences in network structural attributes between social bots and legacy media.Full size tableDiscussionPrimary findingsThis study offers crucial insights into the emerging role of social bots as potential new agenda setters capable of influencing public cognition around major international issues like China’s dual-carbon goals. By integrating NAS theory with computational methods, we provide empirical evidence that social bots can exert stronger agenda-setting influence than legacy media outlets in shaping public cognitive networks during initial periods of salient issue discourse. This finding addresses a key gap in the literature by directly examining social bots’ ability to mold public opinion, expanding our understanding of automated actors as impactful agenda-builders in today’s hybrid media ecosystem.Our MRQAP analysis demonstrates that the influence of social bots and legacy media on public issue networks evolves distinctly over time. During the initial period, social bots exhibited stronger influence compared to legacy media. However, this pattern reversed in the subsequent period, with legacy media’s influence strengthening while bot influence became slightly negative. This suggests that in the early stages of public discourse, the high message frequency and concentrated dissemination patterns of social bots allow them to rapidly set agendas and associate specific concept networks. However, this impact proves relatively short-lived, as legacy media’s influence on the public agenda grows stronger over time. The enduring agenda-setting power of mainstream media is likely rooted in its higher source credibility and ability to shift narratives through sustained coverage as issues evolve. The dissipating agenda-setting influence of social bots over time may be attributable to platform regulatory responses like algorithmic adjustments and account removals once artificial amplification is detected. Additionally, a cognitive elaboration effect (Freelon, 2015) suggests the public becomes increasingly aware and critical of biased narratives, particularly as mainstream media integrates diverse viewpoints.Importantly, the sentiment analysis uncovered distinct emotive stances employed by the different actors—while legacy media maintained a relatively balanced mix of positive, negative, and neutral tones, social bots skewed heavily negative in their sentiment framings regarding China’s climate policies. This negativity bias aligns with theories of automated online propaganda which posit social bots are often purpose-built to provoke outrage, polarize discourse and disrupt constructive engagement on important issues (Aldayel and Magdy, 2022; Boichak et al., 2018; Yuan et al., 2019). Existing literature suggests that social bots’ affective strategies can effectively influence the public’s emotional responses (Zhang et al., 2022). Our results extend this finding by suggesting that emotional framing tactics may be a key mechanism through which social bots initially exert stronger agenda-setting influence compared to legacy media. By injecting a predominance of negative sentiment into environmental discourse, social bots could be shaping the affective trajectories of how public cognition around this issue forms and spreads in early stages. The ability to rapidly disseminate provocative, outrage-inducing emotional frames allows social bots to potentially dominate the associative networks and narratives that first take hold in the public’s mind on emerging issues. In contrast, legacy media’s adherence to norms of impartial and objective reporting (Ojala, 2021) may dictate a more balanced emotional stance that lacks the arousal value to drive agenda-setting until sustained, contextual coverage synthesizes over time.However, our findings indicate this emotive agenda-setting advantage of social bots is relatively short-lived. As issues mature and stabilize beyond initial outrage reaction periods, the credibility and narrative-framing capabilities of mainstream journalism allow legacy media to ultimately regain stronger public agenda-setting power. This suggests the public may experience an affective trajectory from initial bot-induced emotional arousal to a more rationally-grounded understanding shaped by authoritative media framings as events progress.The semantic network analysis further revealed differences in the narratives propagated by each actor. While legacy media coverage emphasizes a diversity of issues and a consensus on cooperation and development, the narrative network of social bots may have more prominently defined climate change centrally as a geopolitical blame game, pinning responsibility on China as the prime environmental polluter. Bots pushed for punitive climate “reparations” rather than emphasizing collaborative solutions. Such dichotomous, accusatory tactics adopted by social bots starkly contrast the balanced journalism norms typically governing environmental coverage, suggesting these automated accounts may in fact be undermining rather than advancing productive societal rhetoric around this consequential issue. In other words, bots’ propensity to promote reductive, zero-sum geopolitical framings inconsistent with neutral positions, which actively distorts and polarizes their discourse rather than enriching public knowledge and debate.Theoretical and practical implicationsOur study makes significant theoretical contributions to the understanding of agenda-setting processes in the contemporary digital media landscape. The research reveals a complex temporal dynamic in how automated and traditional actors compete to shape public discourse. Through empirical analysis using MRQAP techniques, we demonstrate that social bots can temporarily dominate public agenda networks through high-frequency messaging and emotional amplification, before legacy media’s institutional credibility and balanced coverage ultimately prevail. This temporal pattern extends traditional agenda-setting theory by revealing how different communicators exercise influence across the lifecycle of public discourse formation.The findings particularly advance our understanding of the mechanisms through which artificial intelligence agents shape public opinion. By combining sentiment analysis with semantic network examination, we uncover how social bots construct dense, polarized networks optimized for rapid information spread while employing predominantly negative emotional frames. This strategy contrasts sharply with legacy media’s diverse network structures and balanced emotional tone. Such insight enriches Human-Machine Communication (HMC) theory by demonstrating how AI agents construct and propagate cognitive associations in ways fundamentally different from human communicators, while suggesting that enduring agenda-setting power requires more nuanced approaches that build credibility over time.Our semantic network analysis reveals differences in the structure of communication networks between social bots and legacy media. The denser, more tightly clustered networks of social bots suggest a capacity for rapid information dissemination, which may explain their initial influence. However, this structure risks creating echo chambers that may limit long-term impact. In contrast, the more diverse network structure of legacy media may allow for a broader and more sustained influence on public cognition.Moreover, this research extends intermedia agenda-setting theory into new territory by examining the dynamics between human and automated agenda setters. While previous studies have explored relationships between different media types (Guo and Vargo, 2017; McCombs and Shaw, 2005; Su and Borah, 2019), our analysis reveals how AI actors introduce novel patterns to this process. Social bots emerge as a distinct class of communicator capable of temporarily dominating agenda formation through strategic narrative deployment before authoritative media framings take hold. These finding challenges traditional assumptions about intermedia relationships and suggests the need for theoretical frameworks that account for the growing influence of artificial intelligence in public discourse.These theoretical advances carry significant implications for platform governance and environmental communication. Social media platforms need to develop more nuanced approaches to content moderation that consider both the immediate and long-term impacts of bot activity. Rather than simply detecting and removing bots, platforms might focus on understanding how automated accounts influence early discourse formation and develop targeted interventions for critical periods of issue emergence. Environmental and climate change communicators can leverage these insights by developing strategies that acknowledge the different temporal phases of public opinion formation. Understanding how emotional framing and network structures influence public perception at different stages could help create more effective communication approaches that combine immediate impact with sustained engagement.In sum, our research makes significant strides in theorizing the complex dynamics between human and non-human agenda setters in the digital public sphere. The originality of our study lies in its novel examination of the dynamics of social bot influence, its direct comparison of bot and legacy media agenda-setting power, and its innovative integration of sentiment analysis, narrative structure, and with NAS theory. It underscores the need for agenda-setting theories to grapple with the fluid interplay between AI agents, media, and citizens in an era of hybrid human-AI communication. This multifaceted understanding challenges us to move beyond simple dichotomies of “bot vs. human” or “new vs. traditional media” and instead consider how these various forces interact and evolve in shaping public discourse over time. These findings lay the groundwork for future research to further interrogate the evolving mechanisms of agenda-setting in the age of automation, opening new avenues for understanding the intricate relationship between technology, communication, and public opinion formation in our increasingly digitized world.Limitations and suggestions for future researchSome key limitations of this study should be acknowledged. First, the research focuses on the specific context of China’s dual-carbon policy, a unique and multifaceted issue that encompasses climate governance, economic development, and geopolitical considerations. While this case study provides valuable insights into the agenda-setting dynamics surrounding a critical environmental policy, the findings may not be fully generalizable to other public affairs domains. Future research should examine the influence of social bots on public cognitive networks across a broader range of policy issues to establish the robustness and transferability of our conclusions.Second, our study makes a significant advancement over previous Twitter-based research by specifically identifying and analyzing bot-generated content, rather than treating all tweets as a homogeneous dataset (Yao et al., 2022). However, although the bot detection method employed in this study is robust and widely used in the field, it is not infallible. The possibility of misclassifications, whereby some human-operated accounts are mistakenly identified as bots or vice versa, cannot be entirely ruled out. As the sophistication of social bots continues to evolve, future research should strive to integrate more advanced AI techniques, such as those based on deep learning and natural language processing, to improve the accuracy of bot detection (Alothali et al., 2018; Martini et al., 2021). This will help to ensure that the findings accurately reflect the impact of automated actors on public discourse.Third, our methodological approach, while robust, has several important limitations. The MRQAP analysis, though effective for examining network relationships, assumes linear relationships between variables and may not fully capture more complex, non-linear patterns of influence between social bots and public opinion. Additionally, semantic network analysis, while useful for revealing narrative structures, is constrained by its focus on word co-occurrence patterns and may miss subtler aspects of meaning and context. The combination of these methods within the NAS framework also presents limitations in temporal granularity—while we can observe changes between time periods, we cannot track continuous, real-time evolution of agenda-setting influence. Furthermore, computational analysis of social media data faces inherent challenges including potential sampling bias, the difficulty of distinguishing coordinated bot networks from independent bot accounts, and the limitation of analyzing only publicly available content (Van Atteveldt and Peng, 2018). Future research could benefit from mixed-methods approaches that combine our quantitative network analysis with qualitative techniques such as discourse analysis or in-depth interviews. This could provide richer context to our findings, offering insights into how individuals interpret and engage with bot-generated content versus legacy media narratives, and how these interpretations shape public understanding of complex issues like climate policy.Finally, a significant limitation is that we did not differentiate between Chinese and non-Chinese media in our analysis of legacy media influence. This distinction could be crucial, as Chinese and non-Chinese media may have different approaches to covering China’s “dual-carbon” policy, potentially influencing public discourse in distinct ways. Furthermore, this limitation is compounded by the complex regulatory environment surrounding social media in China. The Chinese government’s strict content regulations, implemented through both technological and legal means, significantly shape the landscape of permissible discourse on social media platforms (Tai and Fu, 2020). While our study focused on Twitter, which is officially blocked in China, the government’s information control strategies on domestic platforms like Weibo and WeChat indirectly influence global discourse. This regulatory environment likely affects the types of narratives that can gain traction, even when propagated by social bots on international platforms. Our findings on how social bots shape international perceptions of China’s climate policies should therefore be interpreted within this broader context of governmental influence on public discourse. Future research should strive to distinguish between Chinese and non-Chinese media sources and incorporate a more nuanced understanding of how China’s regulatory environment shapes bot activity and agenda-setting processes, both on domestic platforms and in international social media spaces.In conclusion, as autonomous software agents grow increasingly advanced in their ability to generate human-like persuasive content and influential social networks at massive scale, clearly delineating boundaries between constructive agenda-setting and malicious narrative manipulation by these artificial actors will be critical for preserving integrity in public discourse and democratic decision-making processes. By empirically investigating social bots’ roles as potential new agenda setters versus legacy media, this study represents a vital stride towards mapping the evolving dynamics between media, public, and AI shaping societal cognition in our digitally infused era.
Data availability
The data underlying this article will be shared on reasonable request to the corresponding author.
NotesIt’s important to note that this method doesn’t make judgments based solely on content or AI-generation detection. Instead, it integrates various behavioral and network features to provide probabilistic assessment of bot-like activity.Not all accounts could be successfully analyzed due to various reasons like Twitter’s regular checks and suspensions of bots and other rule-violating accounts, or inability to detect accounts due to Twitter’s ban, deletion of tweets, etc. Approximately 23.1% of the accounts in the dataset could not be analyzed due to such issues.ReferencesAldayel A, Magdy W (2022) Characterizing the role of bots’ in polarized stance on social media. Soc Netw Anal Min 12(1):30. https://doi.org/10.1007/s13278-022-00858-zArticle
PubMed
PubMed Central
MATH
Google Scholar
Alothali E, Zaki N, Mohamed EA, Alashwal H (2018) Detecting social bots on Twitter: a literature review. 2018 International conference on innovations in information technology (IIT), pp 175–180. https://doi.org/10.1109/INNOVATIONS.2018.8605995Anderson JR (1983) The architecture of cognition. Psychology Press. https://doi.org/10.4324/9781315799438Arlt D, Schumann C, Wolling J (2023) What does the public know about technological solutions for achieving carbon neutrality? Citizens’ knowledge of energy transition and the role of media. https://www.h2knowledgecentre.com/content/journal5408Barbieri F, Anke LE, Camacho-Collados J (2021) XLM-T: Multilingual language models in Twitter for sentiment analysis and beyond. arXiv preprint arXiv:2104.12250. https://doi.org/10.48550/arXiv.2104.12250Benham J (2020) Best practices for journalistic balance: gatekeeping, imbalance and the fake news era. Journal Pract 14(7):791–811. https://doi.org/10.1080/17512786.2019.1658538Article
MATH
Google Scholar
Bessi A, Ferrara E (2016) Social bots distort the 2016 US Presidential election online discussion. First monday 21:7–11. https://doi.org/10.5210/fm.v21i11.7090Article
MATH
Google Scholar
Boichak O, Jackson S, Hemsley J, Tanupabrungsun S (2018) Automated diffusion? Bots and their influence during the 2016 U.S. presidential election. In: Chowdhury G, McLeod J, Gillet V, Willett P (eds) Transforming digital worlds. Springer International Publishing, pp 17–26. https://doi.org/10.1007/978-3-319-78105-1_3Broniatowski DA, Jamison AM, Qi S, AlKulaib L, Chen T, Benton A, Dredze M (2018) Weaponized health communication: Twitter bots and Russian trolls amplify the vaccine debate. Am J Public Health 108(10):1378–1384. https://doi.org/10.2105/AJPH.2018.304567Article
PubMed
PubMed Central
Google Scholar
Chadwick A (2017) The hybrid media system: politics and power. Oxford University PressChen C-F, Shi W, Yang J, Fu H-H (2021) Social bots’ role in climate change discussion on Twitter: measuring standpoints, topics, and interaction strategies. Adv Clim Change Res 12(6):913–923. https://doi.org/10.1016/j.accre.2021.09.011Article
MATH
Google Scholar
Davis CA, Varol O, Ferrara E, Flammini A, Menczer F (2016) BotOrNot: a system to evaluate social bots. Proceedings of the 25th international conference companion on world wide web. pp 273–274. https://doi.org/10.1145/2872518.2889302Dekker D, Krackhardt D, Snijders TA (2007) Sensitivity of MRQAP tests to collinearity and autocorrelation conditions. Psychometrika 72:563–581. https://doi.org/10.1007/s11336-007-9016-1Article
MathSciNet
PubMed
PubMed Central
MATH
Google Scholar
Fahmy SG, Abdelgaber KM, Karam OH, Elzanfaly DS (2023) Modeling the influence of fake accounts on user behavior and information diffusion in online social networks. Informatics 10(1):1–27. https://doi.org/10.3390/informatics10010027Article
Google Scholar
Ferrara E, Varol O, Davis C, Menczer F, Flammini A (2016) The rise of social bots. Commun ACM 59(7):96–104. https://doi.org/10.1145/2818717Article
MATH
Google Scholar
Freelon D (2015) Discourse architecture, ideology, and democratic norms in online political discussion. New Media Soc 17(5):772–791. https://doi.org/10.1177/1461444813513259Article
MATH
Google Scholar
Gilardi F, Gessler T, Kubli M, Müller S (2022) Social media and political agenda setting. Politi Commun 39(1):39–60. https://doi.org/10.1080/10584609.2021.1910390Article
MATH
Google Scholar
Guo L (2013) Toward the third level of agenda-setting theory: a network agenda setting model. In: Agenda Setting in a 2.0 World. RoutledgeGuo L, McCombs M (2011) Toward the third level of agenda setting theory: a network agenda setting model. Presented at the Annual Convention of the Association for Education in Journalism & Mass Communication. St. Louis, Missouri. Available at: http://www.leiguo.net/publications/guo_nas_2011_aejmc.pdfGuo L, Vargo C (2015) The power of message networks: a big-data analysis of the network agenda setting model and issue ownership. Mass Commun Soc 18(5):557–576. https://doi.org/10.1080/15205436.2015.1045300Article
MATH
Google Scholar
Guo L, Vargo CJ (2017) Global intermedia agenda setting: a big data analysis of international news flow. J Commun 67(4):499–520. https://doi.org/10.1111/jcom.12311Article
MATH
Google Scholar
Guo L, Vu HT, McCombs M (2012) An expanded perspective on agenda-setting effects. Exploring the third level of agenda setting. Rev Comun 11(1):51–68MATH
Google Scholar
Guo J, Liu S (2022) From# BlackLivesMatter to# StopAsianHate: examining network agenda-setting effects of hashtag activism on Twitter. Soc Media+ Soc 8(4):20563051221146182. https://doi.org/10.1177/20563051221146182Article
Google Scholar
Gokcimen T, Das B (2024) Exploring climate change discourse on social media and blogs using a topic modeling analysis. Heliyon. https://doi.org/10.1016/j.heliyon.2024.e32464Gorwa R, Guilbeault D (2020) Unpacking the social media bot: a typology to guide research and policy. Policy Internet 12(2):225–248. https://doi.org/10.1002/poi3.184Article
Google Scholar
Hallinan B, Kim B, Mizoroki S, Scharlach R, Trillò T, Thelwall M, Segev E, Shifman L (2021) The value(s) of social media rituals: a cross-cultural analysis of New Year’s resolutions. Inf Commun Soc 26:764–785. https://doi.org/10.1080/1369118X.2021.1983003Article
PubMed
PubMed Central
Google Scholar
Holsti OR (1969) Content analysis for the social sciences and humanities. Addison-Wesley, Reading, MAHimelein-Wachowiak M, Giorgi S, Devoto A, Rahman M, Ungar L, Schwartz HA, Epstein DH, Leggio L, Curtis B (2021) Bots and misinformation spread on social media: implications for COVID-19. J Med Internet Res 23(5):e26933. https://doi.org/10.2196/26933Article
PubMed
PubMed Central
Google Scholar
Kim B, Lin H, Kim Y (2024) Interplay of agenda setters in the digital age: the associative issue network between news organizations and political YouTube channels. Comput. Hum Behav 155:108169. https://doi.org/10.1016/j.chb.2024.108169Article
MATH
Google Scholar
Kong F, Wang Y (2022) How to understand carbon neutrality in the context of climate change? With special reference to China. Sustain Environ 8(1):2062824. https://doi.org/10.1080/27658511.2022.2062824Article
MATH
Google Scholar
Li RYM, Wang Q, Zeng L, Chen H (2023) A study on public perceptions of carbon neutrality in China: has the idea of ESG been encompassed? Front Environ Sci 10:949959. https://doi.org/10.3389/fenvs.2022.949959Article
Google Scholar
Lin H, Kim Y (2023) Learning from disagreement on social media: The mediating role of like-minded and cross-cutting discussion and the moderating role of fact-checking. Comput. Hum Behav 139:107558. https://doi.org/10.1016/j.chb.2022.107558Article
MATH
Google Scholar
Lippmann W (1922) Public opinion. RoutledgeLu J-L, Wang C-Y (2018) Investigating the impacts of air travellers’ environmental knowledge on attitudes toward carbon offsetting and willingness to mitigate the environmental impacts of aviation. Transp Res Part D Transp Environ 59:96–107. https://doi.org/10.1016/j.trd.2017.12.024Article
MATH
Google Scholar
Luo H, Meng X, Zhao Y, Cai M (2023) Rise of social bots: the impact of social bots on public opinion dynamics in public health emergencies from an information ecology perspective. Telemat Inform 85:102051. https://doi.org/10.1016/j.tele.2023.102051Article
Google Scholar
Marlow T, Miller S, Roberts JT (2021) Bots and online climate discourses: Twitter discourse on President Trump’s announcement of U.S. withdrawal from the Paris Agreement. Clim Policy 21(6):765–777. https://doi.org/10.1080/14693062.2020.1870098Article
Google Scholar
Martini F, Samula P, Keller TR, Klinger U (2021) Bot, or not? Comparing three methods for detecting social bots in five political discourses. Big Data Soc 8(2):20539517211033566. https://doi.org/10.1177/20539517211033566Article
Google Scholar
McCombs M (2005) A Look at Agenda-setting: past, present and future. Journal Stud 6(4):543–557. https://doi.org/10.1080/14616700500250438Article
MATH
Google Scholar
McCombs ME, Shaw DL (1972) The agenda-setting function of mass media. Public Opin Q 36(2):176–187. https://doi.org/10.1086/267990Article
MATH
Google Scholar
McCombs M, Shaw DL (2005) The agenda-setting function of the press. Oxford University PressMooseder A, Brantner C, Zamith R, Pfeffer J (2023) (Social) media logics and visualizing climate change: 10 years of# climatechange images on Twitter. Soc Media+ Soc 9(1). https://doi.org/10.1177/20563051231164310Mu J, Zhang L, Ma R, Zhang Q, Chen Z, Yang G (2022) A study on mystification strategies of China’s carbon neutrality initiative in american mainstream newspaper. Open Access Libr J 9(5):1–18. https://doi.org/10.4236/oalib.1108786Article
MATH
Google Scholar
Ojala M (2021) Is the age of impartial journalism over? The neutrality principle and audience (dis) trust in mainstream news. Journal Stud 22(15):2042–2060. https://doi.org/10.1080/1461670X.2021.1942150Article
MATH
Google Scholar
Pearce W, Niederer S, Özkula SM, Sánchez Querubín N (2019) The social media life of climate change: platforms, publics, and future imaginaries. WIREs Clim Change 10(2):e569. https://doi.org/10.1002/wcc.569Article
Google Scholar
Posegga O, Jungherr A (2019) Characterizing political talk on Twitter: a comparison between public agenda, media agendas, and the twitter agenda with regard to topics and dynamics. http://hdl.handle.net/10125/59697Rheault L, Musulan A (2021) Efficient detection of online communities and social bot activity during electoral campaigns. J Inf Technol Polit 18(3):324–337. https://doi.org/10.1080/19331681.2021.1879705Article
Google Scholar
Ross B, Pilz L, Cabrera B, Brachten F, Neubaum G, Stieglitz S (2019) Are social bots a real threat? An agent-based model of the spiral of silence to analyse the impact of manipulative actors in social networks. Eur J Inf Syst 28(4):394–412. https://doi.org/10.1080/0960085X.2018.1560920Article
Google Scholar
Santarossa S, Coyne P, Lisinski C, Woodruff SJ (2019) fitspo on Instagram: a mixed-methods approach using Netlytic and photo analysis, uncovering the online discussion and author/image characteristics. J Health Psychol 24(3):376–385. https://doi.org/10.1177/1359105316676334Article
PubMed
Google Scholar
Segev E (2020) Textual network analysis: detecting prevailing themes and biases in international news and social media. Sociol Compass. https://doi.org/10.1111/soc4.12779Shao C, Ciampaglia GL, Varol O, Yang K-C, Flammini A, Menczer F (2018) The spread of low-credibility content by social bots. Nat Commun 9(1):4787. https://doi.org/10.1038/s41467-018-06930-7Article
ADS
CAS
PubMed
PubMed Central
Google Scholar
Shi W, Liu D, Yang J, Zhang J, Wen S, Su J (2020) Social bots’ sentiment engagement in health emergencies: a topic-based analysis of the COVID-19 pandemic discussions on Twitter. Int J Environ Res Public Health 17(22):22. https://doi.org/10.3390/ijerph17228701. ArticleArticle
Google Scholar
Singh R, Zhang Y, Wang H, Miao Y, Ahmed K (2020) Investigation of social behaviour patterns using location-based data—a Melbourne case study. EAI endorsed transactions on scalable information systems. 8(31). https://eudl.eu/doi/10.4108/eai.26-10-2020.166767Su Y, Borah P (2019) Who is the agenda setter? Examining the intermedia agenda-setting effect between Twitter and newspapers. J Inf Technol Polit 16(3):236–249. https://doi.org/10.1080/19331681.2019.1641451Article
MATH
Google Scholar
Su Y, Hu J (2020) A territorial dispute or an agenda war? A cross-national investigation of the network agenda-setting (NAS) model. J Inf Technol Polit 17(4):357–375. https://doi.org/10.1080/19331681.2020.1756553Article
MATH
Google Scholar
Tahamtan I, Potnis D, Mohammadi E, Singh V, Miller LE (2022) The mutual influence of the World Health Organization (WHO) and Twitter users during COVID-19: network agenda-setting analysis. J Med Internet Res 24(4):e34321. https://doi.org/10.2196/34321Article
PubMed
PubMed Central
Google Scholar
Tai Y, Fu KW (2020) Specificity, conflict, and focal point: a systematic investigation into social media censorship in China. J Commun 70(6):842–867. https://doi.org/10.1093/joc/jqaa032Article
MATH
Google Scholar
Van Atteveldt W, Peng TQ (2018) When communication meets computation: opportunities, challenges, and pitfalls in computational communication science. Commun Methods Meas 12(2-3):81–92. https://doi.org/10.1080/19312458.2018.1458084Article
Google Scholar
Vargo CJ (2018) Fifty years of agenda-setting research: new directions and challenges for the theory. Agenda Setting J 2(2):105–123. https://doi.org/10.1075/asj.18023.varArticle
MATH
Google Scholar
Vargo CJ, Guo L (2017) Networks, big data, and intermedia agenda setting: an analysis of traditional, partisan, and emerging online U.S. news. Journal Mass Commun Q 94(4):1031–1055. https://doi.org/10.1177/1077699016679976Article
MATH
Google Scholar
Vargo CJ, Guo L, McCombs M, Shaw DL (2014) Network issue agendas on Twitter during the 2012 U.S. presidential election. J Commun 64(2):296–316. https://doi.org/10.1111/jcom.12089Article
MATH
Google Scholar
Wang Y, Guo C, Chen X, Jia L, Guo X, Chen R, Zhang M, Chen Z, Wang H (2021) Carbon peak and carbon neutrality in China: goals, implementation path and prospects. China Geol 4(4):720–746. https://doi.org/10.31035/cg2021083Article
CAS
MATH
Google Scholar
Woolley SC, Howard PN (2016) Political communication, computational propaganda, and autonomous agents: introduction. Int J Commun 10. https://par.nsf.gov/biblio/10021331-political-communication-computational-propaganda-autonomous-agents-introductionYang K-C, Varol O, Davis CA, Ferrara E, Flammini A, Menczer F (2019) Arming the public with artificial intelligence to counter social bots. Hum Behav Emerg Technol 1(1):48–61. https://doi.org/10.1002/hbe2.115Article
Google Scholar
Yao Q, Li RYM, Song L (2022) Carbon neutrality vs. neutralité carbone: a comparative study on French and English users’ perceptions and social capital on Twitter. Front Environ Sci 10:969039. https://doi.org/10.3389/fenvs.2022.969039Article
Google Scholar
Yuan X, Schuchard RJ, Crooks AT (2019) Examining emergent communities and social bots within the polarized online vaccination debate in Twitter. Soc Media + Soc 5(3):2056305119865465. https://doi.org/10.1177/2056305119865465Article
MATH
Google Scholar
Zeng L, Li RYM, Mao Y, Chen H, Zeng H (2022) A comparative study on LinkedIn and Sina Weibo users’ perceptions of the carbon-neutral city. Front Environ Sci 10:962367. https://doi.org/10.3389/fenvs.2022.962367Article
MATH
Google Scholar
Zhang M, Chen Z, Liu X, Liu J (2024) Theory and practice of agenda setting: Understanding media, bot, and public agendas in the South Korean presidential election. Asian J Commun. https://www.tandfonline.com/doi/abs/10.1080/01292986.2023.2261112Zhang M, Chen Z, Qi X, Liu J (2022) Could social bots’ sentiment engagement shape humans’ sentiment on COVID-19 vaccine discussion on Twitter? Sustainability 14(9):9. https://doi.org/10.3390/su14095566. ArticleArticle
CAS
Google Scholar
Zhang M, Qi X, Liu X, Zhang K (2024) Who leads? Who follows? Exploring agenda setting by media, social bots and public in the discussion of the 2022 South Korean presidential election. SAGE Open 14(2):21582440241248891. https://doi.org/10.1177/21582440241248891Article
Google Scholar
Zhao B, Ren W, Zhu Y, Zhang H (2024) Manufacturing conflict or advocating peace? A study of social bots agenda building in the Twitter discussion of the Russia-Ukraine war. J Inf Technol Polit 21(2):176–194. https://doi.org/10.1080/19331681.2023.2189201Article
Google Scholar
Download referencesAcknowledgementsThis work was supported by the Youth Program of the National Social Science Fund of China (Grant No. 21CXW028), the Major Projects of the National Social Science Fund of China (Grant No. 23&ZD214), and the Youth Program of the National Social Science Fund of China (Grant No. 24CXW045). This work was also supported by a grant from the Postdoctoral Fellowship Program of CPSF (Grant No. GZB20230507).Author informationAuthors and AffiliationsCentre for Chinese Urbanization Studies of Soochow University & Collaborative Innovation Center for New Urbanization and Social Governance of Universities, Suzhou, ChinaHan Lin & Menghan ZhangSchool of Communication, Soochow University, Suzhou, ChinaHan Lin, Menghan Zhang, Xue Qi & Wenqian ShenAuthorsHan LinView author publicationsYou can also search for this author inPubMed Google ScholarMenghan ZhangView author publicationsYou can also search for this author inPubMed Google ScholarXue QiView author publicationsYou can also search for this author inPubMed Google ScholarWenqian ShenView author publicationsYou can also search for this author inPubMed Google ScholarContributionsHan Lin: Formal analysis, Writing—original draft, Writing—review & Editing. Menghan Zhang: Conceptualization, Writing—review & editing. Xue Qi: Data collection, Writing—original draft. Wenqian Shen: Writing—original draft.Corresponding authorCorrespondence to
Menghan Zhang.Ethics declarations
Competing interests
The authors declare no competing interests.
Ethical approval
This article does not contain any research conducted by the authors on human participants, due to the fact that the data analyzed comes from observable data from social media platforms.
Informed consent
This article does not contain any studies with human participants performed by any of the authors.
Additional informationPublisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and permissionsAbout this articleCite this articleLin, H., Zhang, M., Qi, X. et al. Social bots shape public issue networks in China’s dual-carbon agenda: a network analysis using MRQAP.
Humanit Soc Sci Commun 12, 432 (2025). https://doi.org/10.1057/s41599-025-04472-0Download citationReceived: 01 July 2024Accepted: 29 January 2025Published: 26 March 2025DOI: https://doi.org/10.1057/s41599-025-04472-0Share this articleAnyone you share the following link with will be able to read this content:Get shareable linkSorry, a shareable link is not currently available for this article.Copy to clipboard
Provided by the Springer Nature SharedIt content-sharing initiative