Digital diplomacy and domestic audience: how official discourse shapes nationalist sentiments in China

China has increasingly invested in public diplomacy and digital diplomacy to enhance global outreach and defend its discourse power. Assertive messages from the Ministry of Foreign Affairs (MFA) spokespersons circulate not only on foreign platforms (e.g., Twitter) but also domestically via Chinese social media, yet most scholarship focuses on external audiences. This study addresses the relative neglect of how China’s digital diplomacy affects domestic audiences by examining the link between official diplomatic discourse and everyday nationalist sentiments among Chinese netizens. Drawing on Social Identity Theory, the research focuses on two dimensions of nationalist sentiment—national identification (ingroup) and social derogation (outgroup)—and explores how spokespersons’ tones and the salience of external others shape these sentiments. The paper combines discourse analysis, sentiment analysis, and quantitative modeling to test how diplomatic tones influence nationalism directly and indirectly (through identification and derogation), and how these pathways are moderated by the salience of foreign others. The work aims to move beyond crisis-centered studies of nationalism to capture daily, routinized expressions across diverse foreign affairs topics and periods.

The literature on public and digital diplomacy has evolved from a minimalist, foreign-publics focus to recognizing blurred boundaries between foreign and domestic audiences in the digital age. Digital diplomacy is a core component of the “new” public diplomacy, with states engaging varied audiences through online platforms. While studies of China’s digital diplomacy emphasize external impacts—soft power promotion, image-building, and perceptions of “wolf warrior diplomacy”—emerging work highlights domestic implications, including legitimacy-building and representative communications, as diplomatic content is mediated/transmitted to domestic publics. Research on Chinese nationalism includes top-down constructions via elite/media discourse and bottom-up popular nationalism shaping policy preferences; both dynamics interact in a two-way, state–society process. Existing scholarship tends to emphasize high-profile conflicts, potentially biasing our understanding of nationalist sentiments and overlooking daily ‘latent nationalism.’ This article addresses gaps by: (1) examining domestic effects of digital diplomacy on Bilibili, whose bullet screens enable real-time sentiment expression among a large, youth-skewed user base; (2) analyzing a broad array of daily foreign affairs topics beyond major crises; and (3) moving beyond a sole focus on ‘wolf warrior’ offensiveness to consider the broader tonal spectrum of diplomatic discourse. Theoretically, the study links Social Identity Theory to nationalism by distinguishing national identification (ingroup-positive) and social derogation (outgroup-negative), and by positing a key role for the salience of foreign others (especially the U.S.; second tier: Australia, Canada, India, Japan, South Korea, UK) in conditioning comparative dynamics and nationalist expressions.

Design and data: Mixed-method approach combining discourse analysis, automated sentiment analysis, and quantitative modeling. Dataset consists of 200 Bilibili videos of MFA regular press conferences (2019–2022) and their bullet screen comments (122,291 entries). Selection criteria: authenticity of MFA press conference content; ≥50,000 views; ≥100 bullet screens; coverage across diverse themes and countries. Processing and coding: Bullet screens extracted via Python. A nationalism lexicon (544 words) was constructed using NLPIR segmentation; categorized into national identification (225 words) and social derogation (319 words). Gooseeker (automated coding/ML with bag-of-words and neural methods) identified nationalist content and counted category frequencies per video. Percentages of national identification and social derogation were computed relative to total bullet screens per video. Variables:

• Nationalism (outcome): Using BosonNLP sentiment dictionary integrated with the nationalism lexicon, Gooseeker produced sentiment scores on bullet screens with nationalist content. Coding: > +5% = 1 (more positive nationalist sentiments); between −5% and +5% = 0 (mixed/neutral); < −5% = −1 (more negative nationalist sentiments).
• Official diplomatic discourse (ODD, predictor): Spokesperson tone per video coded as −1 (negative: condemnation, sarcasm, refutation), 0 (neutral/mixed: factual/objective), 1 (positive: appreciation, support, welcome). Keywords informed by BosonNLP and the Chinese Diplomatic Language Corpus. Acknowledged precision limits.
• Mediators: Percentages of national identification and social derogation bullet screens per video (identification coded +1 category; derogation coded −1 category for classification purposes).
• Moderator (salience of other): Ordinal coding based on targeted foreign actor: U.S. = 2; Australia, Canada, India, Japan, South Korea, United Kingdom = 1; all others = 0. Analysis: Ordinary least squares (OLS) using PROCESS (Hayes) for mediation, moderation, and moderated mediation. Four-step mediation testing (MacKinnon) with bootstrap confidence intervals (5,000 draws). Robust standard errors clustered by country. Hypotheses: H1 (ODD→nationalism positive); H2 (ODD→nationalism via social derogation); H3 (ODD→nationalism via national identification); H4a/H4b (salience of other moderates these indirect paths).

Descriptive/correlational: Majority of popular nationalist sentiments are positive. Nationalism correlates positively with ODD (r = 0.21, p < 0.01) and national identification (r = 0.58, p < 0.001), and negatively with social derogation (r = −0.71, p < 0.001). Mediation (single-step multiple mediators): Total effect of ODD on nationalism significant (B = 0.24, t = 2.99, p < 0.01; 95% CI [0.08, 0.39]). Total indirect effect significant (0.22; 95% CI [0.08, 0.36]); direct effect small and non-significant (0.02). Indirect via social derogation significant (0.16; 95% CI [0.06, 0.25]) supporting H2. Indirect via national identification not significant overall (0.06; 95% CI [−0.01, 0.15]), providing limited support for H3 in the simple mediation. Moderated mediation: Salience of other significantly moderates both indirect paths (supporting H4a and H4b). For social derogation path (ODD→SD→nationalism): contrast and index of moderated mediation significant (contrast = −0.18, 95% CI [−0.33, −0.03]; index = −0.13, 95% CI [−0.23, −0.02]); stronger inverse ODD→SD effect at higher salience. For national identification path (ODD→NI→nationalism): contrast = −0.22 (95% CI [−0.36, −0.10]); index = −0.16 (95% CI [−0.26, −0.07]); mediation stronger at lower salience levels. Johnson–Neyman intervals indicate the NI-mediated effect is significant only when salience < ~1.14–1.49 (i.e., for salience values 0 or 1), and not present at the highest salience (2). Overall, positive diplomatic tone fosters positive nationalism primarily via increased national identification when the other is less salient; negative tone increases negative nationalism via social derogation, especially when the other is highly salient (e.g., U.S.).

Findings confirm that China’s digital diplomatic discourse, as transmitted to domestic platforms, shapes nationalist sentiments through identity-related mechanisms. Positive spokesperson tones are associated with more positive nationalist expressions via heightened national identification, while negative tones foster negative sentiments through intensified social comparison and outgroup derogation. The salience of foreign others conditions these processes: highly salient others (notably the U.S.) amplify derogation dynamics and diminish identification-based positive mediation, steering sentiments more negative; less salient others allow identification mechanisms to operate, yielding more positive expressions (e.g., pride, inclusion). This refines the understanding of digital diplomacy’s domestic dimension beyond external image management and suggests that assertive rhetoric can elicit nuanced domestic reactions, not solely “hawkish” outcomes. Platform-specific practices matter: Bilibili’s bullet screen culture, youth demographic, and humorous/pop-culture-inflected discourse facilitate real-time, quotidian nationalist expressions, differing from Weibo/Twitter’s formats. The integration of Social Identity Theory offers a multi-faceted view of nationalism, separating ingroup-identification (positive affect) from outgroup-derogation (negative affect), and situating both within the comparative context shaped by salient others.

The study advances theoretical and empirical understanding of how official digital diplomatic discourse influences domestic nationalism in China. It shows that spokespersons’ tones affect nationalist sentiments largely through identity processes—national identification and social derogation—with the salience of foreign others moderating these pathways. Contributions include (1) foregrounding nationalism as a domestic dimension of digital diplomacy; (2) integrating Social Identity Theory to disaggregate nationalist sentiments into identification and derogation; and (3) shifting attention from crisis-driven cases to daily, routinized expressions across varied foreign affairs topics. Practical implications suggest tailoring digital diplomacy strategies and platform-specific messaging to cultivate constructive national identification while mitigating derogation, particularly when addressing highly salient foreign counterparts. Future research should incorporate individual-level demographics, employ more objective tone/salience measures, and explore categorical modeling to enhance robustness and causal inference.

• Lack of demographic/socio-economic data for commenters due to privacy constraints; individual predispositions could not be directly modeled. Future surveys could incorporate demographics and cross-platform comparisons.
• Measurement limitations: tone of official discourse, nationalist sentiment positivity, and salience of others partly rely on subjective coding and automated ML; adding more objective indicators would strengthen validity.
• Variable treatment: ordinal variables (e.g., tone, nationalism, salience) were treated as continuous for OLS and PROCESS simplicity, potentially reducing accuracy; future work can implement categorical models and alternative estimation strategies.