Abstract:

Investigating the differences in emotional perception across residential environments and their underlying mechanisms is essential for enhancing life quality. Using multimodal Weibo UGC data (text and images) from the main urban area of Fuzhou, this study examines emotional perception disparities and environmental impacts among residents in different residential settings. By integrating LSTM-CNN sentiment analysis, XGBoost, SHAP, and ArcMap, we reveal nonlinear effects and variations of residential environments on emotions. The findings indicate: ① Positive emotions account for 68% in formal residential areas versus 83% in urban villages. Formal areas exhibit distinct spatial clustering of emotions with gradual transitions between hotspots and coldspots, whereas urban villages show diffusion patterns, reflecting strong boundary effects and spatial heterogeneity. ② Key positive indicators for urban village residents include paving degree, color complexity, and openness, while formal residential areas are positively influenced by paving degree, visual entropy, spatial vitality, and enclosure. Paving degree and openness exhibit the most significant differential impacts across environments. ③ Excessively high green visibility may be associated with negative emotions. In urban villages, color complexity, visual entropy, and green visibility can mitigate negative emotions induced by enclosed spaces. In formal areas, spatial vitality and enclosure yield more positive outcomes through interaction. This study provides a data foundation for inventory renewal planning in complex residential settings, offers insights into emotional perception differences and environmental mechanisms, and proposes sustainable planning recommendations for enhancing the quality of such environments.

Key words: residential environment, resident emotions, impact mechanisms, machine learning, Fuzhou

摘要：

探究不同住区环境中居民的情绪感知差异与影响机制，对提升居民生活质量具有重要意义。本文以福州市主城区为例，基于图文并行的微博UGC数据，探讨了不同住区环境下居民的情绪感知差异及其环境影响效应，并结合长短期记忆和卷积神经网络（LSTM-CNN）情感分析方法、极度梯度提升树（XGBoost）模型和SHapley Additive exPlanation （SHAP）等机器学习方法，使用ArcMap软件，揭示不同住区环境对居民情绪的非线性影响与差异化作用特征。结果表明：①正式住区居民的积极情绪占比为68%，城中村居民的积极情绪占比为83%。正式住区的情绪空间分布聚集效应显著，冷热点过渡均匀；而城中村则具有扩散性，展现出较强的边界效应和空间异质性。②对城中村居民情绪具有显著正向贡献的指标为：铺装度、色彩复杂度和开阔度；而对正式住区居民情绪的正向贡献指标为：铺装度、视觉熵、空间活力和围合度。在不同住区环境中，铺装度和开阔度的影响差异最大。③过高的绿视率可能引发负面情绪。在城中村提高色彩复杂度、视觉熵和绿视率可以降低围合环境带来的负面情绪影响。同时，正式住区中空间活力和围合度在指标交互时可以产生更多的积极效应。研究结果为复杂住区环境的存量更新规划提供了数据基础，揭示了住区居民的情绪感知差异与环境对情绪的影响机制，为复杂住区环境品质的提升提供了可持续的规划建议。

关键词: 住区环境, 居民情绪, 影响机制, 机器学习, 福州