Social inequality in sight : exploring urban visual perception and sentiment across income levels

Abstract

Visual perception and sentimental expression are interconnected cognitive processes that shape environmental interaction, functioning at different levels across populations. Despite their inherent connection, prior studies have examined them in isolation, focusing on physical environments and overlooking diverse populations. To address this gap, we investigate the connection between urban visual perception and residents' sentiment, with particular attention to income disparities and social inequalities. Deep learning models are used to analyze street view images and social media data to quantify visual perception and sentiment, while separate CatBoost models are trained for each income-level group and compared using interpretability methods. Our findings show that only moderate visual perception enhances sentiment, while low or high perceptual stimulation reduces sentiment. Furthermore, although economically vulnerable residents are exposed to lower-quality perceptions, enhancements in street conditions and income yield greater improvements. These results highlight social inequalities in the urban environment and suggest that unplanned urban sprawl may harm overall well-being.