Multi-Modal Social Media Analysis via SHAP-Based Explanation: A Framework for Public Art Perception

Jiangsu Province’s rapid urbanization presents unprecedented opportunities for examining evolving cultural landscapes through public art perception patterns. We propose a multi-modal framework integrating visual and linguistic analysis: visual content undergoes ResNet-152 feature extraction and Qwen-7B-Chat captioning, while textual data is processed through hierarchical topic modeling and sentiment classification. Through analysis of 103,427 geo-tagged Weibo posts across thirteen administrative regions, this study employs fine-tuned Qwen-7B-Chat architecture integrated with spatial random forest modeling to decode urban-rural variations in artistic appreciation. The methodology reveals counterintuitive relationships between institutional density and cultural reception. Notably, technological firm concentration demonstrates negative correlation with digital art appreciation, while cultural venue density emerges as the dominant predictor for traditional art sentiment. Second-tier cities exhibit sophisticated hybrid indices, with Nantong and Changzhou developing balanced traditional-digital reception patterns independent of metropolitan centers. Educational attainment shows varying impacts across artistic modalities, suggesting differentiated mechanisms of cultural capital formation. These empirical patterns establish a polycentric model of cultural development where mid-sized urban centers cultivate autonomous artistic ecosystems characterized by distinctive institutional-technological configurations. The findings advance theoretical frameworks for understanding cultural diffusion in rapidly urbanizing regions while providing evidence-based recommendations for territorially-calibrated cultural policy implementation.