Portrayals of Chinese companies in American and British economic news tweets during China's macroeconomic transitions 2007–2023

The study examines how portrayals of China-headquartered firms in US and UK economic news tweets vary with China’s changing macroeconomic conditions (2007–2023). Prior work links economic news sentiment to prevailing conditions and negativity bias; however, social media logics and engagement can modulate these patterns. Drawing on van Dijk’s sociocognitive framework, the project treats positive/negative sentiment and emotions as cues of ingroup/outgroup ideologies within media discourse. Because Chinese firms’ performance is influenced by domestic macroeconomic trends, international coverage should reflect such contexts. The study focuses on X (formerly Twitter) given its centrality in disseminating economic news and driving audience engagement via emotive content. Research questions: (RQ1) How do positive/negative sentiments correlate with GDP and PMI, and how do variations shape ideological representation? (RQ2) How do sentiments evaluate companies’ strengths, weaknesses, opportunities, and threats, and with what implications for representation? (RQ3) Which emotions significantly affect sentiment changes and how do they influence representations?

American media have often portrayed China’s economy negatively, especially during crises and trade conflicts, using frames of risk, opposition, and even war metaphors; The New York Times coverage has mixed skepticism with acknowledgment of growth, and US news has emphasized negative depictions of Chinese products during trade disputes. British media display more nuanced, mixed portrayals, balancing risks, ethics, and opportunities: dichotomous representations (market disruptor vs. catalyst), shifting toward risks and human rights after tariff disputes, with recent favorable depictions in some BBC podcast narratives alongside concerns about global trade equity. Sector-specific UK coverage (e.g., 5G/Huawei) often frames Chinese tech firms negatively within geopolitics rather than solely security. Theoretically, the study employs van Dijk’s ingroup/outgroup ideology within a sociocognitive approach: media highlight positive properties of ingroups and negative properties of outgroups through opinions, emotion words, topics, and themes. Prior research used corpus-assisted methods (sentiment analysis, topic modeling, semantic prosody) to analyze portrayals of China’s economy and select firms in traditional media. This study bridges a gap by focusing on social media (X), integrating macroeconomic dynamics with journalistic values to track ideological shifts via targeted sentiments and emotions.

Data collection: Tweets were compiled from 35 verified X accounts of 19 prominent US/UK economic news outlets (e.g., Financial Times, Wall Street Journal), covering financial/business/economic news about Chinese companies from account inception to 12/2023. Using Twint and 22 seed terms (e.g., CHINESE FIRM, frequent Chinese company names), 55,394 tweets (934,155 words; 08/2007–12/2023) were collected and split into US (27,566 tweets; 443,352 words) and UK (27,828 tweets; 490,803 words) subcorpora. Duplicates were retained due to their ideological salience. Chinese quarterly GDP and Composite PMI (manufacturing/non-manufacturing; CEIC) were obtained (08/2007–12/2023). Tweets were categorized into six macroeconomic scenarios based on GDP direction (UP/DOWN) and PMI thresholds (>50, =50, <50): sustained growth, balance, potential decline, potential growth, stagnation, sustained decline. Analytical procedure (four steps): 1) Sentiment and emotion modeling. Two RoBERTa-based models (CardiffNLP) were used via Python: twitter-roberta-base-sentiment (F1=0.76; predicts probabilities of positive/negative/neutral) and twitter-roberta-base-emotion-multilabel (F1=0.72; 11 emotions: anger, anticipation, disgust, fear, joy, love, optimism, pessimism, sadness, surprise, trust). Models output probabilities and labels. In the dataset, 17,074 tweets were positive or negative and 38,320 neutral. 2) Annotation scheme for evaluative targets. Tweets’ thematic targets were defined as: strengths (internal positives: strong financials, brand), weaknesses (internal negatives: financial disappointments, legal infractions), opportunities (external positives: favorable policy, market confidence, IPOs), threats (external negatives: regulation, security designations, market barriers). Rationale derives from business journalism’s coverage of internal performance and external conditions. 3) LLM-assisted annotation. ChatGPT-4 was used (zero-shot prompt engineered with examples and cues incl. economic conditions), with a two-stage process on 17,074 positive/negative tweets. 30% were manually reviewed, yielding intercoder agreement of 75.6%. Disagreements (e.g., separating internal vs external causes) consulted Claude 3 and Microsoft Copilot and were resolved by focusing on causal origin. 4) Statistical analyses. With SPSS 29, Spearman correlations assessed relationships between sentiment probabilities and GDP/PMI (non-normal distributions). Multiple linear regressions (four models for US/UK positivity/negativity) used emotions as predictors of sentiment probabilities, evaluating model fit (R²), multicollinearity (VIF>5 exclusion), 95% CIs (excluding predictors with CIs including zero or larger than coefficients), and significance (p<0.05).

- Significant associations with macroeconomy: Spearman analyses showed significant (p<0.01) correlations between tweet sentiments (N=55,394) and China’s GDP/PMI. As conditions worsened, positivity declined and negativity rose, indicating an ideological shift from ingroup to outgroup portrayals. - US patterns: Positivity peaked at economic balance (GDP UP/PMI=50; median positivity 0.12; median negativity 0.02). During sustained downturns (GDP DOWN/PMI<50), negativity peaked (0.14) and positivity dropped (0.05). US tweets were most positive at balance and most negative during prolonged downturns. - UK patterns: Positivity increased during stability (GDP UP/PMI=50) or potential growth (GDP DOWN/PMI>50) and decreased when a downturn was anticipated (GDP UP/PMI<50). Negativity peaked when a downturn was expected (GDP UP/PMI<50; negativity 0.11; positivity 0.04) and was lower during sustained expansion or potential growth. - Evaluative targets: In positive tweets (N=17,074 pos/neg): strengths and opportunities dominated; in negative tweets, weaknesses and threats dominated. • US positives: Emphasis on high-tech capabilities (smartphones, AI, AV, 5G). Of 3,123 strength-targeted positive tweets, 85% focused on technological advances. Of 1,372 opportunity-targeted positive tweets, 90% highlighted IPOs, inflows, favorable regulation. • US negatives: Of 2,870 weakness-targeted negative tweets, 68% highlighted legal/financial difficulties and competitive challenges. There were 2,031 threat-targeted negatives emphasizing trade tensions, security concerns, and foreign market barriers. • UK positives: Of 2,189 strength-targeted positive tweets, 75% stressed product appeal, market fit, and financial health; opportunities included IPO/share-price gains, regulatory support, and European fundraising access. • UK negatives: Of 2,652 weakness-targeted negative tweets, 65% cited financial shortcomings, safety issues, labor problems, and mismanagement; threats emphasized regulatory actions/bans, antitrust scrutiny, and delisting risks. - Dynamics across scenarios: • US: Strengths were most emphasized at balance (GDP UP/PMI=50) and declined when downturns were anticipated; opportunities increased as the economy deteriorated. Weakness-targeting declined as downturns unfolded, while threat-targeting exceeded weaknesses during decline, peaking at stagnation (GDP DOWN/PMI=50). • UK: Strengths’ share declined with downturns; opportunities increased during recovery (GDP DOWN/PMI>50). Weakness-targeting was highest during recovery and lowest in stagnation; threat-targeting peaked in prolonged recessions (GDP DOWN/PMI<50). - Emotional drivers (multiple regressions; R²>0.53; nine emotions significant at p<0.0001; VIFs acceptable): • US positivity: Joy (B=0.605, p<0.0001) and Trust (B=0.186, p<0.0001) increased positivity; Surprise decreased positivity (B=-0.420, p<0.0001). • US negativity: Fear increased negativity (B=0.202, p<0.0001); Anticipation decreased negativity (B=-0.614, p<0.0001). • UK positivity: Love increased positivity (B=1.510, p<0.0001); Joy increased positivity (B=0.523, p<0.0001). • UK negativity: Fear increased negativity (B=0.185, p<0.0001); Anticipation decreased negativity (B=-0.615, p<0.0001).

The findings address the research questions by demonstrating that tweet sentiment tracks China’s macroeconomic context: positivity rises around stability or growth and falls with deterioration, while negativity expands during anticipated or sustained downturns. These trends align with van Dijk’s ingroup/outgroup ideologies: during favorable conditions, coverage emphasizes ingroup-consistent attributes (strengths and, in some phases, opportunities), whereas worsening conditions elevate outgroup-consistent elements (threats and, at times, weaknesses). Evaluative target shifts show that as macro conditions worsen, positive portrayals pivot from internal strengths to external opportunities (maintaining some ingroup framing), while negative portrayals pivot from internal weaknesses to external threats (deepening outgroup framing). Emotional dynamics further shape these moves: joy, trust, and love bolster positivity and ingroup cues, whereas fear drives negativity and outgroup cues; anticipation mitigates negativity by framing potential rather than realized problems. The cross-national patterns suggest journalistic and market logics: US tweets amplify negativity chiefly during prolonged downturns (potentially balancing investor vigilance with trade continuity), while UK tweets heighten negativity at anticipated downturns (early warning orientation). Overall, the results underscore the interplay of macroeconomic pressures and platform-driven journalistic strategies (engagement via emotions) in steering ideological representations of Chinese firms on social media.

This study integrates macroeconomic indicators with NLP-based sentiment/emotion analysis and LLM-assisted target annotation to reveal how US/UK economic news tweets represent Chinese companies across economic cycles. It shows significant correlations between sentiments and GDP/PMI, systematic shifts from ingroup to outgroup portrayals as conditions worsen, and distinct cross-national timing of positivity/negativity peaks. It contributes methodologically by combining transformer models with LLM-assisted evaluative target annotation and theoretically by linking macroeconomic dynamics to sociocognitive ideological shifts. Future research should widen macro indicators, include additional media systems/regions, and probe causal mechanisms between newsroom practices, platform engagement metrics, and ideological framing.

- Macroeconomic scope: Only GDP and PMI were analyzed, though other indices could add nuance to macroeconomic context. - Media scope: The corpus is limited to US/UK economic news outlets on X; findings may not generalize to other regions or platforms. Future work should include additional indicators and broader international media samples.