Using LLMs to construct location choice models

URBANLY, a company specializing in urban simulation, has integrated large language model approaches into their CityCompass land use simulation platform to improve how they match housing units to people's needs. This new approach aims to better understand how people choose where to live by utilizing LLMs to analyze and incorporate the latest research on household location choices. The system considers various factors like a house's location, size, and nearby amenities, as well as information about the people looking for homes. It then predicts which residential locations different households are likely to choose. This approach opens up new possibilities for incorporating diverse sources of domain knowledge from the literature into land use models. For instance, operators could present a different set of papers to the LLM to explain how households make location choices, and evaluating the forecasts produced by different authors' models and the resulting policy visions.

As tools for rapidly integrating new research findings in household location choice models spread, they have the potentially to accelerate and reshape our understanding and ability to model how people respond to changes in the urban environment taking place due to climate change, enhancing our ability to model and anticipate adaptive behaviors.