Mapping urban emotion from geotagged social media data: Age, gender and spatial heterogeneity

This paper investigates the spatial distribution of emotional expression in urban environments using geotagged social media data, with particular attention to disparities across demographic groups and the contextual factors that shape them. Emotional content is classified into eight categories, derived from text and emojis, using the NRC Emotion Lexicon. A multimodal deep learning model is used to infer the age and gender of the users, allowing the identification of spatial emotion patterns in demographic cohorts. To mitigate compositional bias inherent in social media data, we normalise for demographic representation and analyse within-city variation using geostatistical techniques. Hotspot analyses reveal pronounced spatial disparities in emotional expression by age and gender. Further analysis shows that disparities in emotional expression are significantly associated with environmental exposures (e.g., air pollution, noise levels, heat risk), characteristics of the built environment (e.g. pedestrian and cycling flows), health outcomes (e.g. dementia, obesity, depression prevalence) and behavioural factors such as physical activity and active transport. The findings suggest that unequal exposure to urban conditions shapes differentiated affective experiences, offering new insights into the spatial determinants of subjective well-being. By integrating emotion detection, demographic inference, and spatial modelling, this study provides a scalable and demographically aware framework to analyse affective inequality in cities.