ANNUAL CONGRESS 2025

Emotional Perception Assessment and Influence Factor Identification of Popular Check-in Streets Based on Large Language Models and Interpretable Machine Learning: A Case Study of Old city in Nanjing

Not scheduled

20m

Yildiz Technical University, Istanbul

Poster Track 11 | EMERGING TECHNOLOGIES

Speaker

Mr 松洺 李 (Southeast University)

Description

Urban emotion is a crucial indicator of spatial quality, and emerging popular check-in streets, as new carriers of urban cultural and economic development, deserve particular attention for their emotional characteristics. Previous studies have largely focused on emotional perception at the macro-urban scale, with limited attention to street-level emotional features, while existing evaluation methods struggle to accurately capture multi-dimensional emotional characteristics and their influencing factors. The development of social media data and large language models provides cost-effective data sources and new methods for efficiently identifying subtle emotions.
Using over 500 thousand Weibo posts (China's Twitter) from 2020-2024, this study takes Old City in Nanjing as a case study to develop an emotional perception assessment and influence factor identification model for urban popular check-in streets. The research includes: (1) Identifying high-popularity, high-traffic popular check-in streets through social media data analysis and ArcGIS spatial analysis; (2) Employing advanced large language models with superior text recognition capabilities to quantify citizen-generated content according to Russell's PAD model across three dimensions: Pleasure-displeasure, Arousal-nonarousal, and Dominance-submissiveness, establishing a multi-dimensional emotional perception evaluation model to reveal temporal-spatial patterns of street emotions; (3) Introducing Shapley Additive Explanations (SHAP) as an interpretative model for analyzing potential influencing factors across different emotional dimensions, with cluster analysis for classification.
The study's innovations include: (1) Introducing large language models into street-level emotional perception research, improving emotion recognition accuracy; (2) Establishing a three-dimensional evaluation framework of pleasure-arousal-dominance, enriching street emotion assessment dimensions; (3) Identifying emotional influence factors based on SHAP values, providing data support for street environment optimization. The research findings offer technical support and decision-making reference for urban popular street planning, design, and spatial quality improvement.

References

Yu, Z., Xiao, Z. and Liu, X. (2022) 'A data-driven perspective for sensing urban functional images: Place-based evidence in Hong Kong', Habitat International, 130, p. 102707.

Zhao, X., Lu, Y., Huang, W. and Lin, G. (2024) 'Assessing and interpreting perceived park accessibility, usability and attractiveness through texts and images from social media', Sustainable Cities and Society, 112, p. 105619.

Yin, L., Han, M. and Nie, X. (2024) ‘Unlocking blended emotions and underlying drivers: A deep dive into COVID-19 vaccination insights on Twitter across digital and physical realms in New York, using ChatGPT’, Urban Science, 8(4), pp. 222.