Multi-Modal Social Media Analytics: A Sentiment Perception-Driven Framework in Nanjing Districts

Abstract

This study investigates the complex urban-rural dynamics of Nanjing through a novel multi-modal analysis of 76,288 social media posts, addressing the critical gap in understanding intra-city variations in perceived urban characteristics across rapidly developing Chinese cities. By integrating computer vision techniques with natural language processing, we develop a comprehensive framework for analyzing public sentiment and attention patterns across eleven districts. Our findings reveal distinct spatial gradients in urban perception that challenge traditional urban-rural dichotomies. While central districts exhibit higher positive sentiments toward built environment (0.65-0.71) and economic factors, peripheral areas show stronger positive associations with environmental quality (0.42) and community cohesion (0.47). Correlation analysis demonstrates significant relationships between socioeconomic indicators and digital engagement patterns, with education levels strongly correlating with cultural heritage attention (r=0.76) and income levels with economic discourse (r=0.94). SHAP analysis further reveals non-linear interaction effects between urban characteristics and public sentiment, particularly in transitional zones. These findings contribute to theories of post-reform Chinese urbanization while offering practical insights for targeted urban planning strategies. The study's methodological framework provides a replicable approach for analyzing intra-city variations in urban perceptions across rapidly urbanizing regions.