Using GeoXAI to reveal gender differences in the spatially nonstationary influence of built environment on area-level happiness
- 1College of Art, Wuyi University
- 2College of Landscape Architecture and Art, Fujian Agriculture and Forestry University
Abstract
Gendered activity patterns can create unequal exposure to urban environments, yet city-scale evidence rarely combines gender-specific happiness, nonlinear modelling, and local spatial variation. We analysed 678,120 geotagged Weibo posts across 33,591 500 m grids in Shanghai using LSTM-CNN sentiment classification, HRNet streetscape segmentation, XGBoost, SHAP, partial dependence plots, and spatially varying contributions. After screening, 320,618 female-labelled and 187,089 male-labelled posts entered the analysis. Sentiment classification achieved 0.942 accuracy; random-fold XGBoost R ² was 0.898 and 0.930 for female- and male-labelled happiness, and remained positive under spatial separation and the 20-user threshold. Results showed: (1) positive expression accounted for 73.64% of female-labelled posts and 74.30% of male-labelled posts. Female-labelled grid scores were higher and more variable (4.06 ± 2.13 versus 2.94 ± 1.71); both surfaces were spatially clustered (Moran’s I = 0.909 and 0.666) but modestly correlated ( r = 0.325; ρ = 0.273). (2) Population count and the Gini coefficient showed opposite correlations between groups. SHAP ranked NDVI first for the female-labelled model and Population count first for the male-labelled model, with Greenness, the Gini coefficient, and Distance to city center also prominent. (3) The same 5D dimension dominated only 43.3% of paired grids. Destination accessibility and Distance to Transit contributed more to female-labelled happiness in 95.1 and 84.0% of grids, whereas Design contributed more to male-labelled happiness in 68.0%. All five dimension-level differences were spatially clustered (Moran’s I = 0.571–0.788). These results locate contiguous gender differences in area-level happiness across Shanghai and provide evidence for locally targeted, gender-responsive urban and public-health planning.
Related
- The effect of differences in viewing environment and method on physiological responses and perceptual stability while viewing three-dimensional images
- Emergency Department or Ward? Does the Insertion Environment Influence Bloodstream Infection Risk After Central Venous Catheter Placement?
- Reference method matters: DXA system differences influence agreement with 3D optical imaging body composition estimates
- The relationship between aging in place and happiness levels among older adults: A cross-sectional study
- Effects of a people care-based song-centered group music program on perceived parental care, self-esteem, peer relationships, and happiness in school-age children
BibTeX
@article{ye2026using,
title = {{Using GeoXAI to reveal gender differences in the spatially nonstationary influence of built environment on area-level happiness}},
author = {Cuicui Ye and Zheng Ding},
journal = {Frontiers in Public Health},
year = {2026},
month = oct,
doi = {10.3389/fpubh.2026.1962036},
url = {https://www.frontiersin.org/articles/10.3389/fpubh.2026.1962036/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Using GeoXAI to reveal gender differences in the spatially nonstationary influence of built environment on area-level happiness
AU - Cuicui Ye
AU - Zheng Ding
PY - 2026
DA - 2026/10/14
JO - Frontiers in Public Health
DO - 10.3389/fpubh.2026.1962036
UR - https://www.frontiersin.org/articles/10.3389/fpubh.2026.1962036/full
AB - Gendered activity patterns can create unequal exposure to urban environments, yet city-scale evidence rarely combines gender-specific happiness, nonlinear modelling, and local spatial variation. We analysed 678,120 geotagged Weibo posts across 33,591 500 m grids in Shanghai using LSTM-CNN sentiment classification, HRNet streetscape segmentation, XGBoost, SHAP, partial dependence plots, and spatially varying contributions. After screening, 320,618 female-labelled and 187,089 male-labelled posts entered the analysis. Sentiment classification achieved 0.942 accuracy; random-fold XGBoost R ² was 0.898 and 0.930 for female- and male-labelled happiness, and remained positive under spatial separation and the 20-user threshold. Results showed: (1) positive expression accounted for 73.64% of female-labelled posts and 74.30% of male-labelled posts. Female-labelled grid scores were higher and more variable (4.06 ± 2.13 versus 2.94 ± 1.71); both surfaces were spatially clustered (Moran’s I = 0.909 and 0.666) but modestly correlated ( r = 0.325; ρ = 0.273). (2) Population count and the Gini coefficient showed opposite correlations between groups. SHAP ranked NDVI first for the female-labelled model and Population count first for the male-labelled model, with Greenness, the Gini coefficient, and Distance to city center also prominent. (3) The same 5D dimension dominated only 43.3% of paired grids. Destination accessibility and Distance to Transit contributed more to female-labelled happiness in 95.1 and 84.0% of grids, whereas Design contributed more to male-labelled happiness in 68.0%. All five dimension-level differences were spatially clustered (Moran’s I = 0.571–0.788). These results locate contiguous gender differences in area-level happiness across Shanghai and provide evidence for locally targeted, gender-responsive urban and public-health planning.
SN - 2296-2565
ER -