Alibakhshi,R., & Srivastava, S. C. (2025).

Journal of the Association for Information Systems,26(4), 1081-1137.

‍

This research analyzes 24,000 Instagram posts to examine “sentiment divergence”—mismatch between image and caption sentiments—and its impact on engagement. Drawing on expectation-disconfirmation and negativity-bias theories, the authors show that greater divergence reduces likes and comments; moreover, negative divergence (caption more negative than the image) is especially harmful, while equally large positive divergence is less damaging (pp. 2–4, 9–10). A second contribution introduces “duality tolerance”—a macro-cultural trait—showing that communities more comfortable with contradictions (e.g., Eastern cultural contexts) experience weaker engagement penalties from divergence, highlighting the importance of audience culture when crafting multimodal posts (pp. 4, 6–7, 10). Methodologically, the study operationalizes image sentiment via Microsoft Azure facial-emotion cues and text sentiment via text2emotion, then computes divergence magnitude and direction; extensive controls (e.g., face area, adult/racy/gore scores, sentiment complexity) bolster robustness (pp. 11–14). Managerially, recommendations include aligning emotional tone across visuals and captions, avoiding unexpectedly negative captions under upbeat imagery, and localizing creative to the cultural “duality tolerance” of follower communities. The paper advances influencer research beyond single-mode sentiment by modeling cross-modal affective coherence as a determinant of engagement and by coupling content analytics with cultural orientation—providing a practical, testable framework for post design and audience targeting.