Reevaluate Chinese online political discourse: Nationalism VS. rationalism

The digital transformation of political discourse has fundamentally altered the conditions under which public discussion occurs, intensifying a core tension between rational deliberation and emotional mobilization. Habermas [27]’s foundational conception of the public sphere as a domain of rational-critical debate encounters new complexities in digital environments, where algorithmic amplification tends to privilege emotionally charged, out-group hostile content over nuanced argumentation [28]. This tension between rational and emotional discourse is not unique to deliberate democracy contexts—even in China’s authoritarian system, scholarship remains divided on whether nationalism serves as the dominant ideological force or whether digital platforms create space for more pluralistic political discourse.

Several strands of scholarship on digital nationalism converge in showing how Chinese online spaces, especially in the context of Sino–Japan relations, systematically transform historical grievance into contemporary nationalist mobilization through platform dynamics. Schneider’s study of the Nanjing Massacre shows how state control over digital infrastructure and algorithmic curation produces a highly homogeneous nationalist narrative, in which historical memory is continuously reactivated and stabilized across websites, encyclopedias, and hyperlink networks [29]. Building on this infrastructural account, Ng and Han demonstrate how everyday platform interactions further intensify this dynamic: in responses to the Japanese celebrity Sora Aoi on Weibo, nationalist antagonism toward Japan becomes intertwined with misogynistic abuse, revealing how affective and victimized nationalism is enacted through personalized and emotionally charged discourse rather than formal political argument [30]. Together, these findings resonate with Guo, Cheong, and Chen’s notion of “latent nationalism,” in which platform architectures and attention economies allow nationalist sentiment to be triggered with minimal deliberation [31]. Taken as a whole, this literature helps explain how environmental controversies involving Japan become rapidly nationalized in Chinese digital spaces, as technical risks are framed into historical injury and amplified through affective, algorithmically mediated public engagement.

Conversely, a competing body of scholarship challenges this nationalism-dominant narrative, arguing instead that Chinese digital spaces foster more pluralistic political discussions than is often assumed. Liu [32]’s qualitative analysis of online debates during the 2005 anti-Japan protests revealed that cyber nationalism functions not merely as an extension of state propaganda, but as a counterforce that creates space for negotiating national identity beyond official narratives. Although extreme nationalism exists, the online environment allows for debate and dissent that complicate simple state-society binaries. This finding echoed with Zhang [25]’s quantitative analysis, which demonstrated that Chinese online political discourse was not dominated by nationalistic views, but rather by pro-democracy perspectives that explicitly criticize government policies. These divergent findings indicate that existing scholarship may benefit from moving beyond the binary question of whether nationalism or pluralism dominates Chinese digital discourse. A more nuanced approach would examine how nationalist and rationalist modes of discourse interact and compete within the same digital spaces.

The Fukushima wastewater release provides an ideal case for examining how nationalist and rationalist discourse compete in Chinese digital spaces. This issue combines technical scientific questions about radiation safety and cosmopolitan concerns over global environment, alongside deep nationalist sensitivities regarding Japan’s historical role in China. Drawing on Habermas [27]’s concept of communicative rationality, which emphasizes reasoned argumentation aimed at mutual understanding, in this study we define rationalism as a mode of discourse grounded in evidentiary deliberation—scrutinizing claims through scientific data and logical coherence—and transnational norms, appealing to ethical frameworks that extend beyond parochial nationalism.

Our study aims to reveal how internet users navigate between nationalistic responses and rational discussions. Thus, the following research question is posited:

RQ1: How does the relative prevalence of nationalism versus rationalism discourse shift throughout different phases of the Fukushima wastewater release?

Shaping cyber nationalism: Key opinion leaders (KOLs) VS. grassroots

The Fukushima wastewater release sparked both nationalist discussions and rational responses. While rational discourse offers an evidence-based perspective, nationalistic discourse is assumed to dominate. In recent years, Chinese online space has witnessed the marginalization of rational and moderate voices, a structural trend shaped by rigorous institutional regulations [33], algorithmic amplification of affective content [13], and sophisticated propaganda which is constantly adapting to new digital affordances [34–36]. Given this structural landscape, our study, while acknowledging the presence of rational discourse, focuses primarily on the dynamics of nationalist discourse.

If the rise of the internet initially led to a weakening of state control over public expressions of national belonging and “turned the hitherto state-centered logic of nationalism on its head” [37], the growing monopoly of platforms has enabled the state to reassert control over national imagination, while also opening doors for other political and corporate actors to interfere in the process [13]. This transformation has empowered multiple actors—state institutions, Key Opinion Leaders (KOLs), and grassroots users to shape nationalistic discourse within an increasingly complex digital ecosystem.

State-led nationalism has received considerable attention among scholars. Gries [38] argued that regime legitimacy is the key mediator accounting for the association and sequencing between popular nationalism and China’s Japan policy. Cairns [39] uncovered how government censorship on Weibo fluctuated throughout the Diaoyu/Senkaku dispute, reflecting a strategic response to manage nationalistic sentiment and prevent social instability. State-controlled nationalism exerts significant influence by shaping the information infrastructure through “techno-nationalism” and propaganda [40,41]. However, the direct mechanisms of state intervention often remain opaque due to undisclosed decision-making in authoritarian nations, making it difficult to measure their precise impact. Therefore, this study treats official nationalism as a structural backdrop, instead focusing on the observable interactions between KOLs and grassroots users.

Within the Chinese social media ecosystem, key opinion leaders (KOLs) serve as influential intermediaries in shaping public opinion. Weibo endorses these KOLs through verification badges, marking them as “Big Vs” with substantial follower bases and social influence [42]. Unlike paid subscription models on platforms such as X (formerly Twitter), Weibo enforces a real-name authentication system where verification is contingent upon professional credentials and quantitative influence metrics, including specific requirements for follower counts and readership. These verified accounts mainly fall into three categories: official media outlets representing party-state interests with diplomatic tones [43]; commercial “We-Media” (Zimeiti) accounts that align ideological correctness with “ideotainment”—a blend of propaganda and popular culture to maximize traffic [44,45]; and public intellectuals—such as former Global Times editor Hu Xijin—who offer nuanced policy interpretations while adhering to party-state interests [42]. Their verified status and large followings make them effective channels for top-down communication of state-sanctioned narratives. This influence was evident in Guo [46]’s analysis of the 2014 Hong Kong Umbrella Movement discussions, where mainland Chinese netizens echoed these opinion leaders’ paternalistic attitudes and nationalistic rhetoric, framing the protests as an anti-China conspiracy with Western backing.

Meanwhile, grassroots users have demonstrated an increasing agency in shaping nationalist narratives, although the nature of this mobilization has undergone a critical transition. Early studies highlighted the bottom-up dimensions of nationalistic expression, which were typically precipitated by specific geopolitical crises [47]. An example is the 2016 Diba Expedition to Facebook, where Taiwan presidential election triggered a voluntary online campaign through popular forms of Chinese memes [48]. However, recent scholarship suggests that this agency has evolved from reacting to clear political disputes to a hyper-sensitive vigilance known as “Ruhua” (辱华) or “insulting China” accusations. Wu and Zhang [49] argue that recent popular nationalism has shifted to a rigid defense of state dignity, which can be incited by seemingly innocuous content—such as a casual photo or a sports itinerary—provided it can be interpreted as challenging the state’s territorial integrity or historical narratives. This trend indicates that threshold for “insulting China” is lowered, allowing unrelated cultural slights to rapidly escalate into crises of national dignity.

Despite these valuable insights, the nuanced roles among various actors in co-constructing online nationalist discourse remain underexplored. Most studies focus on either top-down state influence or bottom-up grassroots participation, but rarely examine how different social actors construct and potentially diverge in their nationalist interpretations within the same discursive space. While verified KOLs typically amplify state narratives, grassroots users represent bottom-up participation in nationalist discourse. This study addresses this gap by examining the dynamic interplay between these two actors in constructing nationalist narratives, using the Fukushima wastewater release as a case study, leading to the following question:

RQ2: What are the relative roles of grassroots users and KOLs in shaping nationalist narratives during the Fukushima event?

Data collection and preprocessing

Data collection utilized the open-source GitHub tool dataabc/weibo-search, retrieving posts through two keywords: nuclear wastewater (核废水) and nuclear polluted water (核污水), commonly used in Chinese discourse and official reports. Original posts (原创微博) were crawled from July 4, 2023, to March 18, 2024, encompassing key events such as the IAEA’s assessment of Japan’s ocean discharge plan on July 4, 2023, Prime Minister Fumio Kishida’s announcement on August 22, 2023, and the completion of the fourth discharge round on March 17, 2024. A total of 70,759 original posts were collected, along with metadata including hashtags, verification status, and postdate.

Although each collected post contained at least one of the target keywords, a subset of the data contained noise that did not reflect genuine user opinions on Fukushima wastewater release. We manually excluded these non-substantive posts, which included: reposts from third-party platforms (e.g., Douyin, Little Red Book); lottery notifications (identified by the token sequence “Lottery Details”); posts consisting solely of hashtags; and unrelated topics ranging from entertainment news and investment advice to civil service exam tips that utilized Fukushima-related hashtags to increase visibility, a practice known as “hashtag hijacking.” While we acknowledge that keyword-based filtering has its limitations, efforts were made to filter out posts that did not genuinely reflect user opinions on the Fukushima wastewater issue. After filtering, 56,797 posts remained. Table 1 presents the distribution of Weibo posts after filtering by user type, showing that ordinary accounts contributed the majority of posts (71.16%, n = 40,416) compared to verified accounts (28.84%, n = 16,381) out of a total of 56,797 posts. Table 2 shows the top 10 Chinese hashtags found in posts after filtering that matched the keyword search, along with their English translations, drawn from a total of 12,732 instances of hashtags.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Distribution of Weibo Posts by User Type.

https://doi.org/10.1371/journal.pone.0352452.t001

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Top 10 Most Frequent Hashtags Extracted from Collected Weibo (n = 12,732).

https://doi.org/10.1371/journal.pone.0352452.t002

We subsequently cleaned the text by removing structural noise, including username mentions, Weibo video tags, URLs, and “Super Topics” (超话). Super Topics are interest-based community tags on Weibo, mainly used for celebrity fandoms and gaming. As these tags represent platform-specific metadata rather than substantive discourse regarding the wastewater incident, they were stripped to prevent semantic noises. Finally, stopwords were removed and text segmentation was performed using the Jieba package designed for Chinese text. To ensure the topic model captured distinct thematic variations rather than the common denominator of the dataset, the search keywords themselves (“nuclear wastewater” and “nuclear contaminated water”) were treated as corpus-specific stop words and removed during tokenization.

Measuring dynamic discourse with structural topic modeling (STM)

In this study, Structural Topic Model (STM) was chosen due to its unique advantages. One of the primary benefits of employing STM is its ability to incorporate document-level metadata (covariates) to influence both topic prevalence and content [50]. This feature distinguishes STM from traditional topic modeling methods such as LDA, which typically utilize Gibbs sampling or variational inference for estimating model parameters and focus solely on word distributions [51].

We aimed to examine the evolution of discourses as the discharge process progresses and differential issue attention between key opinion leaders and grassroots, the date of the posts and user type emerged as crucial covariates. To analyze the relationship between our chosen covariates and topic prevalence, we employed a regression-based approach within the STM framework. The date of the posts was modeled using a smooth function spline to capture potentially non-linear temporal effects on topic prevalence, allowing for flexible modeling of time-varying effects without imposing a rigid linear structure [52]. User type, categorized as key opinion leaders (KOLs) or grassroots, was included as a categorical covariate. We employed a global approach to uncertainty quantification, ensuring robust and conservative estimates [53], and estimated the individual and interaction effect of covariates separately.

To prepare the textual data for STM, we utilized the textProcessor function. A critical parameter in this step was the determination of token length. To align with the linguistic characteristics of Chinese, where unigrams are often functional or ambiguous, and substantive meaning is encoded in compound words, we filtered out all tokens consisting of a single character, ensuring that the model was constructed exclusively on multi-character tokens. It resulted in the removal of 271 documents that no longer contained valid text tokens, yielding the final analytical corpus of 56,526 documents reported in our results.

As STM is an unsupervised model, another crucial initial step is determining the optimal number of topics to represent the text corpus. To achieve this, searchK function was utilized to compute diagnostic values for models with different values of K. Our topic number selection followed a two-step validation process to balance statistical rigor and interpretability, leading to the choice of 7 as the optimal number of topics due to lower residuals and higher semantic coherence. Firstly, we conducted a wide-range screening (K = 5–50) as Fig 1 shows, where statistical fit metrics showed only marginal improvements as the number of topics increased. The stabilization of the lower bound beyond K = 10 indicated diminishing returns [54]. Then, we performed a narrow-range selection (K = 5–10) as Fig 2 shows. The Ideal solutions are characterized by fewer residuals and higher exclusivity, variational lower bound, and semantic coherence [55]. Our aim was to derive actionable insights rather than corpus-specific optimization, making semantic coherence the most crucial metric due to its strong correlation with human judgment () and its precedence over likelihood in applied settings [56]. K = 7 was selected as the optimal number of topics because it demonstrated favorable semantic coherence (−88.31) while maintaining low residuals (6.99). Models with K > 7 exhibited a substantial deterioration in semantic coherence, with K = 8 dropping to −97.27 and K = 10 plummeting to −99.81—representing a 10% and 13% decrease respectively from K = 7. Although K = 10 achieved slightly lower residuals (6.56), this marginal improvement was overshadowed by the dramatic loss in semantic interpretability, making K = 7 the optimal choice that balances model fit with meaningful topic coherence.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Wide-Range Number of Topics Screening (K = 5–50).

Structural topic model diagnostics (held-out likelihood, semantic coherence, residuals, and variational lower bound) surveyed over a wide topic range (K = 5–50) to enable broad exploration and identification of a plausible region for the optimal number of topics.

https://doi.org/10.1371/journal.pone.0352452.g001

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Narrow-Range Number of Topics Selection (K = 5–10).

Structural topic model diagnostics (held-out likelihood, semantic coherence, residuals, and variational lower bound) evaluated over a deliberately narrow topic range (K = 5–10) to enable fine-grained, precise selection of the optimal number of topics.

https://doi.org/10.1371/journal.pone.0352452.g002

Two antagonisms in nationalistic discourse: Anti-Japanese and Anti-Western

To analyze the discourse patterns in the dataset, we employed a mixed-method approach combining computational topic modeling and qualitative coding. We generated a Structural Topic Model (STM) from 56,526 documents using a dictionary of 4,461 terms, resulting in 7 distinct topics (Fig 3). Manual coding categorized these topics based on nationalist and rationalist discourses. Nationalist discourse, marked by anti-Japanese and anti-Western sentiments (Table 3), appeared in Topics 6 (Anti-Japanese) and 7 (Anti-Western). In contrast, rationalist discourse, characterized by evidence-based arguments and global humanitarian concerns, was found in Topics 2 (Environmental & Ecological Concerns) and 3 (Discharge Standards). Topics 1, 4, and 5 were classified as ‘other/mixture,’ covering market impacts, regional responses, and sparse personal reactions. This hierarchical analysis improved interpretability and facilitated theoretical saturation evaluation [57]. Fig 4 presents a word cloud of terms associated with nationalistic topics (Topics 6 & 7).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Manual Categorization Based on Structural Topic Modeling Result.

https://doi.org/10.1371/journal.pone.0352452.t003

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Highest Word Probability for Each Topic.

Top-probability words for each topic in the 7-topic structural topic model of Chinese texts. Bars represent highest estimated token probabilities ().

https://doi.org/10.1371/journal.pone.0352452.g003

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Nationalism Discourse Wordcloud.

Chinese-language word cloud of Anti-Japanese (Topic 6) and Anti-Western (Topic 7) Weibo posts. Token size is proportional to relative frequency after basic preprocessing. High-frequency tokens concentrate on: National actors and identity (中国 China, 日本 Japan, 民族 nation, 我们 we, 美国 USA) and emotive geopolitical critique (抵制 boycott, 鬼子 derogatory anti‑Japanese term, 战争 war).

https://doi.org/10.1371/journal.pone.0352452.g004

The first layer of antagonism shows how historical grievances serve as a foundation 331 for anti-Japanese sentiment:

#Japan’s nuclear wastewater# I suggest that our country (China) should report more related news. This thing is much scarier than COVID-19 and its impact on humanity will last for generations. The Japanese appear polite on the surface, but in private, they are still the same Japanese devils. The atrocities of Unit 731 have not been brought to justice, and now there is nuclear wastewater. Everyone should condemn Japan’s discharge, for the sake of the Earth, for all living beings on Earth, and for all humanity.

In Topic 6, the antagonistic logic primarily draws upon historical trauma, particularly referencing World War II atrocities (“Japanese devils” and “Unit 731”). This discourse strategically juxtaposes Japan’s contemporary actions with historical grievances, suggesting a continuity of malevolent intent beneath a veneer of civility (“appear polite on the surface”). In this context, the controversial global environmental issue is reframed as symbolic of Japan’s historical disregard for global well-being—an act rooted in the same moral failings that characterized its wartime atrocities. This narrative also implies a victimhood mentality by portraying China as a moral guardian standing against Japan’s recurring harm to humanity.

The second layer indicates a conspiracy-oriented antagonism that extends beyond bilateral tensions to implicate the broader Western power structure:

Tokyo Electric Power Company in Japan is actually an American company! I’ve always wondered why Japan insists on discharging nuclear wastewater into the ocean, which harms others without benefiting itself. Meanwhile, the United States and Europe, who usually champion environmental protection, have collectively remained silent, and even the International Atomic Energy Agency has been bought off. Just look at the major shareholders of Tokyo Electric Power Company! The largest shareholder of Tokyo Electric Power Company is the renowned Wall Street giant BlackRock, which is also a major financier of the Ukrainian comedian. The second-largest shareholder, Vanguard, is also a Wall Street giant, both are Jewish capital……

In Topic 7, the conspiracy theories about Western financial institutions reflect how external threats are perceived as coordinated actions by Western powers. The terms “Jewish capital” and “US-Japan Water” reflect an antagonistic worldview among Chinese internet users that frames global politics through a US-China antagonistic lens, categorizing the international community into two camps: ‘China and its allies’ and ‘the US and its allies.’ Within this framework, the IAEA’s authorization is interpreted as a manifestation of how the US-Japan alliance operates to compromise China’s national interests, implying a sense of collective victimhood which positions China as a target of systemic Western collusion.

Overall, the STM identifies two dominant nationalist antagonisms—anti-Japanese (Topic 6) and anti-Western (Topic 7)—that structure how the Fukushima wastewater issue is discussed online. At the evaluative level, Topic 6 frames Japan’s discharge through historical grievance and moral condemnation, linking contemporary environmental risk to wartime trauma, while Topic 7 interprets the issue through conspiracy-oriented geopolitical narratives that position Western actors as coordinated threats to China. Together, these topics reframe a technical environmental controversy as a nationalist narrative centered on collective victimhood and moral opposition, demonstrating how environmental discourse is systematically absorbed into broader patterns of digital nationalist antagonism.

Dynamics of nationalism and rationalism discourses across key phases

Fig 5 visualizes the time-varying proportions of nationalism and rationalism topic categories.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Topic Proportion Trends Over Time.

It shows how topic proportions between nationalism and rationalism fluctuate throughout the observed period. The solid lines represent the expected proportions of each topic, while the dotted lines denote the 95% confidence intervals. Visually, periods where confidence intervals do not overlap suggest separation in the prevalence of the respective topics. Conversely, overlapping confidence intervals suggest insufficient evidence to assert a visual separation. Formal inference relies on the model-based estimates reported in the text.

https://doi.org/10.1371/journal.pone.0352452.g005

Phase 1: The Intensification of Nationalism (Early July 2023 – Early August 2023)

The first phase illustrates a dynamic interplay between nationalism and rationalism, with rational discourse (Topics 2 and 3) exceeding nationalist topics (Topics 6 and 7) during the critical window of July 4–11, 2023, coinciding with the IAEA’s safety review release. Rationalism initially dominated, peaking at a combined proportion of 0.41 on July 4. Statistical divergence occurred between July 11 and July 14; the proportion of Topic 3 declined sharply, while Topic 6 exhibited an inflection point, rising from 0.18 to 0.45. By July 14, the 95% Confidence Intervals (CIs) of Topic 6 [0.207, 0.225] and Topic 3 [0.145, 0.162] ceased to overlap, confirming a statistically significant transition (p < .05) toward nationalist framing. This shift aligns with the immediate aftermath of the IAEA’s safety review, suggesting that technical assurances acted as a brief stimulus for rational debate before being eclipsed by grassroots nationalist mobilization.

Phase 2: Stabilization and Diffusion (Mid-August 2023 – January 2024)

Seven days right before Japan’s official discharge commencement (August 24, 2023), rational concerns resurged briefly, with Topic 2 (Environmental & Ecological) peaking at 0.192 (95% CI [0.189, 0.196]) on August 17—a last gasp of technical discourse before nationalist narratives solidified. Nationalism retained structural dominance compared with rationalism, as the combined proportion of Topic 6 and Topic 7 is larger than that of Topic 2 and Topic 3 across this phase.

Phase 3: Resurgence of Nationalism (February 2024)

A dramatic nationalist resurgence occurred in February 2024, with Topic 6 (Anti-Japanese) surging to 0.379 (95% CI [0.373, 0.386]) on Feb 18—its highest level since August 2023. This was triggered by two events: A technical mishap (40-minute Fukushima leak on February 7), reigniting distrust in Japan’s stewardship, and the “Messi incident” (February 10), where the athlete’s perceived snub of Hong Kong in favor of Japan amplified anti-Japanese sentiment.

Fukushima nuclear wastewater leak is due to forgetting to close the manual valve. It’s not forgetting to close the manual valve, it’s forgetting the basics of being human. Japan is doing things that are not humane!

#Why Did Messi Only Respond to Missing The Game After Arriving In Japan? # Because of the small island’s nuclear waste treatment.

In Chinese cyberspace, the actions of foreign celebrities are frequently viewed through a nationalistic lens, where gestures can be interpreted as signs of either respect or disrespect toward China. The controversy surrounding Lionel Messi exemplifies this dynamic: his absence from a Hong Kong match and subsequent visit to Japan sparked widespread accusations of “insulting China” (辱华), with internet users criticizing his apparent preference for Japan. This public reaction suggests that strong national pride may transform perceived slights into feelings of victimization. Despite China’s growing global influence, online discourses appear to remain highly sensitive to actions that are perceived as prioritizing other nations – particularly Japan – over China. This sensitivity manifests in collective indignation, often expressed through satirical comments, as seen when netizens linked Messi’s behavior to Japan’s controversial nuclear wastewater discharge. Such responses arguably reflect a deeper underlying narrative. As China achieves greater economic and cultural milestones, there is an expectation of corresponding international recognition. This external validation is crucial for putting the “Century of National Humiliation” (百年国耻) firmly in the past. Therefore, any perceived disrespect challenges this modern narrative of national rejuvenation. It may easily serve to reawaken deep-seated historical feelings of victimization.

These findings complicate the notion that Chinese social media discourse is uniformly nationalistic. While nationalist narratives ultimately dominated discussions of Japan’s Fukushima wastewater release, the initial phase (July 4–11, 2023) revealed a coexistence of rationalism and nationalist perspectives. During this critical window, technical concerns about discharge standards (Topic 3) and environmental impacts (Topic 2) competed vigorously with anti-Japanese (Topic 6) and anti-Western (Topic 7). But the balance proved fragile as nationalistic discourse, especially anti-Japanese (Topic 6) became mainstream. Notably, anti-Western sentiment (Topic 7) maintained a stable presence (0.06–0.16 proportions) even as anti-Japanese rhetoric fluctuated wildly. This stability may suggest that anti-Western narrative operate as a stable structural feature of Chinese digital nationalism, whereas anti-Japanese sentiment functions as a situational amplifier activated by specific triggers. The latter phenomenon—where a sports figure’s itinerary reignited historical animosities—exemplifies what we term latent nationalist sensitivity: a psychological state where perceived offense to China rapidly reactivates historical victimhood, transforming trivial incidents into crises of national dignity. This mechanism, while resonating with earlier findings of latent nationalism [31], exemplifies a unique characteristic of Chinese digital nationalism, where “insulting China”(辱华) narratives are rapidly amplified through online discourse.

The rise of grassroots nationalism in Sino-Japan conflict

Fig 6 shows the static estimated mean difference in topic proportion between KOLs and grassroots, where the difference is calculated as (verified account – unverified account).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 6. Estimated Effect of User Type on Topics.

The x-axis represents the difference in topic prevalence, where positive values indicate higher prevalence among KOLs and negative values indicate higher prevalence among grassroots. Each point on the plot represents the estimated mean difference for a topic, with horizontal lines showing the 95% confidence intervals; if a confidence interval does not cross zero, it suggests a statistically significant difference in topic prevalence between the two user types for that topic.

https://doi.org/10.1371/journal.pone.0352452.g006

Raw, volatile nationalism targeting Japan was primarily a bottom-up phenomenon as grassroots users significantly dominated the Topic 6 (Anti-Japanese) discourse (Effect Size = −0.089, 95% CI [−0.093, −0.085]). While grassroots users also led the discussion on Topic 7 (Anti-Western), the gap between the two groups was narrower (Effect Size = −0.026, 95% CI [−0.029, −0.022]). This suggests that although KOLs largely avoided direct anti-Japanese antagonism, they were more aligned with the general public regarding broader anti-Western geopolitical framing. In contrast, KOLs functioned as anchors for diplomatic and rational discourse, showing the most substantial positive preference for Topic 4 (Responses from Neighboring Countries) (Effect Size = 0.144, 95% CI [0.140, 0.149]) and Topic 3 (Discharge Standards) (Effect Size = 0.080, 95% CI [0.077, 0.083]).

Fig 7 shows a more detailed and time-specific view, revealing that these differences were not consistent across time. Notably, the relationship between grassroots and KOLs is not a simple binary of top-down orchestration versus bottom-up expression, but rather depends on the specific target of nationalism.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 7. Temporal analyses on topic proportions across KOLs and grassroots.

The figure provides temporal expected topic proportions grouped into three categories(nationalism, rationalism and other/mixture) between KOLs and grassroots. Blue denotes KOL; red denotes grassroots accounts. Solid lines show posterior mean expected topic proportions; dotted lines trace the upper and lower 95% confidence interval bounds.

https://doi.org/10.1371/journal.pone.0352452.g007

For anti-Japanese discourse (Topic 6), the divergence between grassroots and KOLs remained robust throughout the entire observation period. Grassroots users took a significantly higher proportion of attention to this topic than KOLs, with the difference remaining statistically significant at all time points. Even at the point of closest convergence on February 5, 2024, a clear gap persisted: grassroots mean was 0.295 (95% CI: [0.287, 0.302]) compared to the KOL mean of 0.265 (95% CI: [0.254, 0.276]). This indicates that while the intensity of the gap fluctuated, KOLs constantly maintained a lower profile toward direct anti-Japanese rhetoric compared to grassroots users.

However, with anti-Western discourse (Topic 7), the gap vanished during initial stage of discharge. Specifically, from July 17 to August 14, 2023, the difference in expected topic proportions became statistically insignificant. Within this window, particularly on July 30, KOLs exhibited a higher mean proportion (0.088, 95% CI: [0.074, 0.101]) than grassroots users (0.082, 95% CI: [0.077, 0.088]). Unlike the persistent stratification seen in anti-Japanese discourse, anti-Western discourse saw a temporary synchronization where state-affiliated attention aligned with grassroots sentiment during certain period.

These findings complicate the binary paradigm of Chinese digital nationalism as either purely state-orchestrated or entirely bottom-up, pointing instead to a dynamic and nuanced interplay between the state and populace. While grassroots users drove the volatile anti-Japanese sentiment (Topic 6), the relative silence of Key Opinion Leaders (KOLs) may suggest a degree of strategic restraint, permitting organic expression without explicit official endorsement. However, the discursive synchronization observed in anti-Western discourse (Topic 7) suggests that state-aligned KOLs may opportunistically leverage the public’s victimhood-centric worldview at certain junctures. Together, these patterns indicate that rather than uniformly dictating nationalist expression, the state apparatus may effectively modulate it by selectively channeling grassroots sentiment to support broader geopolitical objectives, while preserving diplomatic flexibility.