Global policy shocks trigger complex waves of public sentiment that vary across regions and social groups. However, the mechanism by which regional economic sensitivities and social influence structures jointly shape these responses remains poorly understood. Focusing on United States tariff policies in early 2025, this study examines how event driven shocks across various sectors trigger sentiment shifts and information diffusion within the Weibo ecosystem. We constructed a large scale dataset comprising 1.3 million posts, including 230,000 geolocated entries. To enable high precision analysis, we implemented a weakly supervised learning framework by fine tuning a specialized RoBERTa model using pseudo labels generated by DeepSeek-v3. Our analytical framework integrates temporal dynamics, Geographically Weighted Regression (GWR) to capture spatial heterogeneity, and information cascade modeling to evaluate network diffusion. The results reveal a structured societal response. Temporally, sentiment follows a multi peak evolution across three strategic policy stages, where event cycles are dynamically defined using moving average smoothing and a 20\% peak intensity threshold. Spatially, GWR identifies a critical polarization phase where local economic factors explain up to 57.2\% of sentiment variance. Topologically, we observe a structural divergence in propagation: institutional accounts (Blue V) drive explosive breadth, whereas elite influencers and regular users act as the primary drivers of multi level cascade depth. These findings quantify how digital public opinion is co-constructed by geographic economic realities and social network hierarchies.
Gli shock legati alle politiche globali scatenano complesse ondate di opinione pubblica, ma l'interazione tra sensibilità economiche regionali e strutture di influenza sociale è ancora poco compresa. Focalizzandosi sulle politiche tariffarie degli Stati Uniti all'inizio del 2025, questo studio esamina come gli shock guidati dagli eventi in vari settori inneschino cambiamenti nel sentimento e la diffusione delle informazioni all'interno dell'ecosistema di Weibo. Abbiamo costruito un set di dati su larga scala di 1,3 milioni di post, inclusi 230.000 record geolocalizzati, e implementato un framework di apprendimento debolmente supervisionato perfezionando un modello RoBERTa specializzato tramite etichette generate da DeepSeek-v3. L'analisi integra dinamiche temporali, regressione geograficamente pesata (GWR) per catturare l'eterogeneità spaziale e modellazione delle cascate di informazioni per valutare la diffusione di rete. I risultati rivelano una risposta sociale strutturata. Temporalmente, il sentimento segue un'evoluzione a picchi multipli attraverso tre fasi politiche strategiche, con cicli di eventi definiti dinamicamente tramite medie mobili e una soglia di intensità del 20\% rispetto al picco. Spazialmente, la GWR identifica una fase di polarizzazione critica in cui i fattori economici locali spiegano fino al 57,2\% della varianza del sentimento. Topologicamente, emerge una divergenza strutturale nella propagazione: gli account istituzionali (Blue V) guidano l'ampiezza esplosiva, mentre gli influencer e gli utenti comuni sono i motori primari della profondità delle cascate multilivello.
Spatio-temporal dynamics and network diffusion of Weibo sentiment toward U.S tariff policies: a geographic perspective
DENG, JIANWEI
2025/2026
Abstract
Global policy shocks trigger complex waves of public sentiment that vary across regions and social groups. However, the mechanism by which regional economic sensitivities and social influence structures jointly shape these responses remains poorly understood. Focusing on United States tariff policies in early 2025, this study examines how event driven shocks across various sectors trigger sentiment shifts and information diffusion within the Weibo ecosystem. We constructed a large scale dataset comprising 1.3 million posts, including 230,000 geolocated entries. To enable high precision analysis, we implemented a weakly supervised learning framework by fine tuning a specialized RoBERTa model using pseudo labels generated by DeepSeek-v3. Our analytical framework integrates temporal dynamics, Geographically Weighted Regression (GWR) to capture spatial heterogeneity, and information cascade modeling to evaluate network diffusion. The results reveal a structured societal response. Temporally, sentiment follows a multi peak evolution across three strategic policy stages, where event cycles are dynamically defined using moving average smoothing and a 20\% peak intensity threshold. Spatially, GWR identifies a critical polarization phase where local economic factors explain up to 57.2\% of sentiment variance. Topologically, we observe a structural divergence in propagation: institutional accounts (Blue V) drive explosive breadth, whereas elite influencers and regular users act as the primary drivers of multi level cascade depth. These findings quantify how digital public opinion is co-constructed by geographic economic realities and social network hierarchies.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_3_Deng .pdf
accessibile in internet per tutti
Dimensione
10.18 MB
Formato
Adobe PDF
|
10.18 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/252680