Background: Influenza seasons may be associated with increased anxiety-related expressions on social media. Social media can reflect population-level emotional expression patterns in real time.

Objective: The aim of the study is to characterize diurnal and full-season dynamics of anxiety-related language during the 2023-2024 influenza season in China and its association with influenza activity.

Methods: We retrieved Sina Weibo posts in February 2025 covering September 4, 2023, to April 28, 2024. Posts containing Diagnostic and Statistical Manual of Mental Disorders (DSM)-based anxiety terms were cleaned and deduplicated (N=169,728 → 106,440). We first linked weekly influenza incidence with anxiety-related postings. Then, we plotted diurnal patterns by epidemiologic phase, conducted supplementary within-sample hourly normalization analyses, and modeled longitudinal symptom trajectories using ARIMA (autoregressive integrated moving average) and supplementary ARIMAX (autoregressive integrated moving average with exogenous regressors) time-series models.

Results: Anxiety-related posts closely followed influenza activity, surging during the outbreak and peak phases and remaining elevated even after influenza declined. Initial Spearman correlation analyses showed significant negative associations for irritability (r=-0.413; P=.02) and restlessness or feeling keyed up or on edge (r=-0.396; P=.02). However, supplementary ARIMAX analyses further revealed that being easily fatigued and difficulty concentrating or mind going blank exhibited more stable positive temporal associations with influenza activity after controlling for autocorrelation and lagged effects. Diurnal patterns shifted across stages, showing mild early-evening variation during the outbreak, clear morning peaks with secondary afternoon and evening rises during prevalence and decline, and morning-afternoon concentration in the end stage. Supplementary within-sample hourly normalization analyses showed that the major diurnal structures remained generally stable after normalization. ARIMA time-series analysis revealed that irritability and being easily fatigued consistently dominated the discussions, whereas others remained at relatively low levels. Out-of-sample forecasting based on a chronological 80% training and 20% testing split suggested generally stable short-term trajectories, with being easily fatigued showing a slight increase.

Conclusions: This study demonstrates how social media can capture diurnal and seasonal fluctuations of anxiety symptoms associated with influenza activity, advancing understanding of affective dynamics in population health.