LLMS para extração de características visuais de áreas urbanas: avaliação e aplicação no contexto de resultados eleitorais

Abstract

The growing political polarization observed globally has highlighted the need to understand the factors that influence electoral results. Although it is widely known that social, racial and economic aspects shape the political positions of individuals, the impact of the physical political positions of individuals, the impact of the physical environment on electoral polarization is still a little explored field. Explored. This study proposes an innovative methodology based on Large Language Mo- dels (LLMs) to extract physical characteristics from urban images that favor social interaction. Interaction. Based on this analysis, we classify urban areas according to their potential for social interactions and investigated the relationship between this characteristic and electoral electoral results in the city of Chicago, USA. Our results suggest a possible positive influence between the degree to which an area favors social interaction and a large electoral advantage for a given candidate in the same area. Electoral advantage of a given candidate in the same region. This approach may offer new valuable insights into other factors that influence electoral outcomes, complementing traditional analyses based on socioeconomic and demographic variables. This study paves the way for future investigations into how urban planning and the physical physical configuration of cities can impact not only social dynamics, but also the political behavior of communities. Political behavior of communities.