Studying the impact of emotion-AI in cross-cultural communication on the effectiveness of global media

The study examines how emotion-AI affects cross-cultural communication in global media, addressing a noted gap wherein prior research focused largely on technical aspects or localized contexts rather than worldwide communicative impact. Drawing on high-/low-context cultural theory, the Technology Acceptance Model (TAM), and emotion as social information (EASI) perspectives, the paper positions emotion-AI as a means to bridge cultural divides, enhance emotional connection, and improve understanding across diverse audiences. It underscores the need for culturally adaptive, emotionally accurate AI systems in media production and distribution. The work advances a novel methodological approach—fsQCA—to capture complex interrelations among user experiences, perceptions, and reactions, aiming to clarify how and why emotion-AI influences global media effectiveness.

The paper situates emotion-AI within communication theories and cross-cultural frameworks, referencing high-/low-context culture (Afrouzi, 2021) and EASI (Van Kleef et al., 2010; van Doorn et al., 2015) to emphasize the importance of context, nonverbal cues, and culturally mediated emotional interpretation. It leverages TAM and related adoption frameworks (Na et al., 2022; Wang C. et al., 2023; Ho et al., 2024) to explain user acceptance based on perceived usefulness and ease of use. Prior studies highlight the promise of emotion recognition in media engagement (Madanian et al., 2023; Wang J. Z. et al., 2023; Ballesteros et al., 2024) while cautioning against universalist models that may misread culturally specific expressions (Barrett, 2017; Barrett et al., 2019; Pugh et al., 2021; Khare et al., 2024). The review identifies a gap in examining emotion-AI’s global communicative effects across cultures and motivates the need for culturally informed algorithmic design and datasets.

Design: Descriptive cross-sectional survey targeting global media consumers exposed to emotion-AI. The design supports capturing relationships among experiences, perceptions, and reactions at a single time point. Population and Sampling: Convenience sampling via social media and digital channels; N=1,108 respondents. Gender: 50.2% female (556), 49.8% male (552). Age: 18–29 (40.9%), 30–39 (22.6%), 40–49 (16.8%), below 18 (8.4%), 50–59 (5.9%), 60+ (5.5%). Education: Bachelor's (64.6%), Secondary (19.2%), Master's (12.6%), Doctorate (3.5%). Media exposure frequency indicated substantial engagement (daily 36.3%, weekly 28.2%, occasionally 19.9%, frequently 11.8%, never 3.9%). Data Collection: Structured online survey administered via Wenjuanxing (wjx.cn). Instrument sections: demographics; emotional experiences with emotion-AI in media; perceptions of reliability and cultural significance; reactions to emotion-AI’s application in communication. Likert scales (1–5). Pilot-tested for clarity and reliability. Model and Variables: fsQCA framework with REACTIONS = f(PERCEPTIONS, EXPERIENCES). Dependent variable: favorable reactions to emotion-AI. Independent variables: EXPERIENCES (quantity/quality of interactions) and PERCEPTIONS (reliability and cultural significance). Calibration and Analysis: Data calibrated to fuzzy-set membership values using minimum, mean, and maximum thresholds (Experiences: min 1.4, mean 3.730686, max 5; Perceptions: min 1.6, mean 3.718412, max 5; Reactions: min 1.6, mean 3.745126, max 5). Necessity analysis tested single conditions and their negations; truth table constructed to assess combinations of EXPERIENCES and PERCEPTIONS. Intermediate solution derived using fsQCA 3.0 with frequency cutoff 176 and consistency cutoff 0.863592. Key metrics: Experiences consistency 0.858083, coverage 0.852179; Perceptions consistency 0.858083, coverage 0.864803. Solution coverage 0.928306, solution consistency 0.81264.

Exposure and Experiences: Approximately 63% reported frequent encounters with emotion-AI-mediated content (mean 3.77, SD 1.114), indicating widespread presence in global media. Perceived Accuracy and Cultural Value: About 60% agreed emotion-AI accurately interprets emotions across cultures (mean 3.69, SD 1.142). Around 63% agreed emotion-AI improves understanding of diverse cultural perspectives (mean 3.76, SD 1.123) and enhances emotional connection in cross-cultural communication (~62%, mean 3.72, SD 1.167). Ease of engagement was positively rated by ~61% (mean 3.72, SD 1.145). Perceptions of Role: Roughly 61% agreed emotion-AI bridges cultural communication gaps (mean 3.72, SD 1.156) and is reliable for complex emotions (~61%, mean 3.71, SD 1.135). About 61% saw emotion-AI as providing meaningful insights into cultural nuances (mean 3.73, SD 1.118). A concern regarding oversimplification was noted (34.7% worried), indicating limitations in capturing nuanced cultural manifestations. Reactions and Outlook: Confidence in handling cultural diversity was ~63.2% (mean 3.73, SD 1.114). 61.6% agreed emotion-AI enhances global media effectiveness (mean 3.72, SD 1.132). 62.3% agreed it makes content more engaging and relatable (mean 3.74, SD 1.137). Optimism about evolution to address limitations was highest at 63.4% (mean 3.8, SD 1.121). ~62% viewed emotion-AI as a necessary innovation (mean 3.72, SD 1.134). fsQCA Results: Necessity tests showed positive EXPERIENCES and PERCEPTIONS meet the ≥0.85 consistency threshold, indicating they are necessary conditions for favorable reactions (Experiences consistency 0.858, coverage 0.852; Perceptions consistency 0.858, coverage 0.865). Truth table combinations with Experiences=1 and Perceptions=1 had highest raw consistency (0.919) and largest case count (401). Intermediate solution achieved solution coverage 93% and consistency 81%, evidencing that positive experiences and perceptions together are sufficient pathways to positive reactions.

Findings address the research question by demonstrating that emotion-AI can strengthen cross-cultural communication in global media through increased exposure, perceived cultural insight, and enhanced emotional connection. The fsQCA results clarify that favorable reactions hinge on two necessary conditions—positive experiences and positive perceptions—underscoring the importance of culturally adaptive design and emotional accuracy. The study situates these outcomes within TAM and EASI, highlighting perceived usefulness/ease of use and culturally contingent emotion interpretation. It also identifies oversimplification risks and the need for richer, culturally diverse training data and context-aware algorithms. The discussion advocates integrating cognitive science concepts such as Theory of Mind and metacognition to improve contextual sensitivity, error monitoring, and adaptive responses, thereby reducing misinterpretation in high-context cultures and expanding effectiveness across diverse communicative norms.

The study provides empirical evidence that emotion-AI, when perceived as reliable and culturally meaningful and experienced positively by users, can enhance global media communication by bridging cultural gaps and fostering emotional connections. Using fsQCA, it demonstrates that experiences and perceptions are necessary drivers of favorable reactions, achieving high solution coverage and consistency. Contributions include addressing a literature gap on emotion-AI’s global communicative impact, introducing fsQCA to this domain, and offering design implications for culturally adaptive, emotionally accurate systems. Future research should pursue iterative algorithmic refinement, diversify training datasets, and embed Theory of Mind and metacognitive mechanisms to improve contextual understanding and self-monitoring. Longitudinal and cross-cultural comparative studies, interdisciplinary collaborations, and exploration of synergies with generative AI and advanced NLP are recommended to further strengthen emotion-AI’s role in inclusive, effective cross-cultural media communication.

The study relies on a cross-sectional, convenience sample collected via online channels, which may limit causal inference and generalizability across populations. Self-reported Likert responses are subject to response biases. Demographic skew toward highly educated participants may affect representativeness. Methodologically, fsQCA calibration choices and cutoffs may influence solution pathways. Substantively, current emotion-AI systems exhibit oversimplification risks and difficulty capturing nuanced, culture-specific nonverbal cues, potentially constraining interpretation accuracy in high-context settings.