Spatial-Knowledge-Graph-Grounded LLM Agents for Neighborhood Livability Evaluation

This research introduces a prototype framework leveraging spatial knowledge graphs (KGs) and large language models (LLMs) to enhance neighborhood livability evaluation. Unlike traditional static indicators, this approach aims to capture the dynamic, individual experiences of residents by generating and revising household schedules, followed by feasibility checks and GIS-based network materialization. This methodology integrates diverse data points, including residents, facilities, and neighborhood context, to provide a more nuanced understanding of how different individuals interact with their environment.

Why it matters

This development is strategically important as it offers a novel, dynamic method for assessing urban environments, moving beyond static metrics to understand how diverse populations experience their surroundings. This shift enables more data-driven decision-making in urban planning, resource allocation, and policy development, potentially leading to more equitable and efficient service provision and infrastructure investments.

Key insights

• Traditional neighborhood livability assessments often rely on static built-environment indicators that do not fully capture resident experiences.
• The proposed framework uses spatial KGs and LLMs to create dynamic, individualized representations of neighborhood interaction.
• It generates and revises household schedules, considering varying mobility capacities, household roles, schedules, and care responsibilities.
• Rule-based feasibility checking and GIS-based network materialization are integrated to validate these dynamic assessments.
• A spatial KG structure integrates diverse elements such as residents, residences, facilities, neighborhood context, and road networks.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.25952