Geographic AI for Understanding Urban Housing and Neighborhoods: Elizabeth Delmelle

Elizabeth Delmelle, Associate Professor City and Regional Planning and director the Master of Urban Spatial Analytics (MUSA) program (University of Pennsylvania) Description: In this talk, I will present two case studies that leverage recent advances in AI-enabled Natural Language Processing and Large Language Models to study urban housing markets. In the first, I will introduce word embeddings to classify a large, temporal corpus of real estate listings in Charlotte, North Carolina, creating an unsupervised classification of how agents have characterized neighborhoods over time. We use an artificial neural network, the self-organizing map, to visualize this high-dimensional space and trace neighborhood pathways to construct a longitudinal typology of change. We find that the primary pathway of change is a shift toward investor-oriented marketing language in older suburban neighborhoods. In the second example, we generate LLM (Claude) perceptions of Chicago neighborhoods to explain neighborhood premiums. In this case, the words are incorporated into a regression model along with more traditionally captured quantitative indicators used to explain neighborhood values – like census variables, proximity to amenities, and measured crime rates. We find that AI-generated text captures nearly as much variation in neighborhood premiums as conventional measures. Sponsored by the Interdisciplinary Innovation Grant on GIS and Spatial Visualization Register via LibCal to get the Zoom link: https://libcal.brynmawr.edu/event/17849783