CityLLM: A framework for natural-language querying of semantic 3D city models

2026-07-15

AI Summary

This study addresses the challenge faced by non-expert users and interdisciplinary researchers in efficiently accessing semantically rich yet structurally complex 3D urban models formatted in specialized standards. To bridge this gap, the authors propose the first lightweight and extensible conversational query framework that integrates large language models—such as GPT-OSS, Gemini 3.1, and GPT-5.4—with spatial and graph databases to automatically translate natural language queries into structured database commands and generate corresponding visualizations. The framework supports four types of complex queries: spatial, graph-based, cross-database, and multi-turn dialogues. Evaluated on 54 test cases, it achieves answer accuracy rates of 85.2%–100%, visualization correctness of 92.9%–100%, and successfully completes all queries with an average of fewer than three retries.

Abstract

Semantic 3D city models provide rich geometric and semantic information, but remain challenging for non-experts and interdisciplinary researchers to access and query due to their complex structures and specialized data formats. To address this issue, we present CityLLM, a framework for natural-language querying of semantic 3D city models alongside complementary urban datasets. The framework combines spatial and graph databases within an LLM-based workflow that supports iterative query refinement and cross-database chaining. We evaluate CityLLM on a CityJSON dataset of Rotterdam (853 LoD2 buildings) using GPT-OSS, Gemini 3.1, and GPT-5.4, along with selected variants, across multiple metrics: answer correctness, visualization correctness, query success, and retry attempts. A total of 54 natural-language queries are curated across four scenarios: spatial, graph, cross-database, and conversational. Results show strong overall performance, with answer correctness ranging from 85.2% to 100%, visualization correctness from 92.9% to 100%, a 100% query success rate, and fewer than three retries across all 54 queries. Overall, the findings suggest that CityLLM provides a lightweight and extensible approach for conversational access to semantic 3D city data.

Problem

• semantic 3D city models
• natural-language querying
• accessibility
• interdisciplinary research
• complex data formats

Innovation

• natural-language querying
• semantic 3D city models
• LLM-based framework
• spatial-graph database integration
• conversational GIS