• Assess walkability characteristics using ChatGPT and Google Street View images. • ChatGPT generates walkability scores and narrative descriptions of walkability issues. • Apply natural language processing to analyze ChatGPT-generated descriptions. • Identify detailed walkability issues and their spatial patterns. • Validate ChatGPT’s evaluations with human ratings and computer vision models. A better understanding of walkability in minority neighborhoods requires close attention to route quality , which includes diverse streetscape features, but comprehensive large-scale assessment remains challenging. This study introduces a novel approach that integrates ChatGPT with Google Street View (GSV) to conduct scalable, fine-grained walkability evaluations in Los Angeles, California. For each GSV image, ChatGPT generates both numeric walkability scores and narrative descriptions of negative aspects. LDA topic modeling is used to uncover latent walkability topics in these narratives. Spatial analyses of both outputs reveal that minority neighborhoods consistently exhibit lower walkability scores and face disproportionate challenges, such as unkempt streetscapes. Validation against human ratings and computer vision models is conducted to assess the reliability of ChatGPT-based evaluations. The findings demonstrate that ChatGPT can capture nuanced microscale features and social cues beyond the capabilities of existing off-the-shelf computer vision methods. This approach provides a context-rich, scalable tool for targeted and equity-focused interventions in minority neighborhoods.
Ki et al. (Tue,) studied this question.