The Emotional Valence Paradox in Social media: A Computational Linguistic Analysis of How Sentiment, Emotion, and Content Domain Predict Social Comparisons on Rednote

ABSTRACT This computational linguistic study analyzes 8743 Rednote posts to examine how sentiment, emotion, and content domain collectively shape social comparison dynamics on social media. We identify four key findings: (1) an emotional gradient, where upward comparison content exhibits the highest sentiment scores and downward comparison the lowest; (2) an emotional valence paradox, wherein positively valenced upward content may trigger psychological harm (e.g., anxiety, reduced self‐esteem) via upward comparison, while negatively valenced downward content may impose affective labor burdens (e.g., pity, empathy) via downward comparison; (3) bidirectional emotion‐comparison relationships, where specific emotions (e.g., joy, disgust) simultaneously trigger and result from comparison processes; and (4) content domain predictors, with education‐related content associated with downward comparisons and political content with upward comparisons. Anchored in the cultural discourse of involution, these findings challenge conventional assumptions about emotional valence in digital communication, revealing how positive sentiment in competitive ecosystems may inadvertently cause harm. We propose the sentiment‐emotion‐semantics (SES) perspective to explain how content features relationally shape comparison orientations. The study provides implications for emotion‐aware platform designs to mitigate performative vulnerability, media literacy programs to address valence paradoxes, and policies to reduce comparison‐induced harms. These insights advance both social comparison theory and computational linguistic approaches to digital affect.