Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OGD4All: A Framework for Accessible Interaction with Geospatial Open Government Data Based on Large Language Models

Michael Siebenmann, Javier Argota Sánchez-Vaquerizo, Stefan Arisona, Krystian Samp, Luis Gisler, Dirk Helbing

TL;DR

The paper addresses the challenge of enabling citizens to interact with geospatial Open Government Data through trustworthy, semantically grounded interfaces. It introduces OGD4All, a two-stage framework that combines semantic dataset retrieval, iterative Python-code generation for geospatial analysis, and secure sandboxed execution to produce verifiable multimodal outputs, all within an auditable workflow. On a Zurich-specific benchmark of 199 questions across 430 datasets and 11 LLMs, the approach achieves high analytical correctness (approximately ninety-eight percent) and recall (approximately ninety-four percent) while reliably rejecting unsupported queries to minimize hallucinations. The work emphasizes transparency-by-design, reproducibility, and governance accountability, and highlights the need for sovereign/open models to balance performance with public trust and data sovereignty.

Abstract

We present OGD4All, a transparent, auditable, and reproducible framework based on Large Language Models (LLMs) to enhance citizens' interaction with geospatial Open Government Data (OGD). The system combines semantic data retrieval, agentic reasoning for iterative code generation, and secure sandboxed execution that produces verifiable multimodal outputs. Evaluated on a 199-question benchmark covering both factual and unanswerable questions, across 430 City-of-Zurich datasets and 11 LLMs, OGD4All reaches 98% analytical correctness and 94% recall while reliably rejecting questions unsupported by available data, which minimizes hallucination risks. Statistical robustness tests, as well as expert feedback, show reliability and social relevance. The proposed approach shows how LLMs can provide explainable, multimodal access to public data, advancing trustworthy AI for open governance.

OGD4All: A Framework for Accessible Interaction with Geospatial Open Government Data Based on Large Language Models

TL;DR

The paper addresses the challenge of enabling citizens to interact with geospatial Open Government Data through trustworthy, semantically grounded interfaces. It introduces OGD4All, a two-stage framework that combines semantic dataset retrieval, iterative Python-code generation for geospatial analysis, and secure sandboxed execution to produce verifiable multimodal outputs, all within an auditable workflow. On a Zurich-specific benchmark of 199 questions across 430 datasets and 11 LLMs, the approach achieves high analytical correctness (approximately ninety-eight percent) and recall (approximately ninety-four percent) while reliably rejecting unsupported queries to minimize hallucinations. The work emphasizes transparency-by-design, reproducibility, and governance accountability, and highlights the need for sovereign/open models to balance performance with public trust and data sovereignty.

Abstract

We present OGD4All, a transparent, auditable, and reproducible framework based on Large Language Models (LLMs) to enhance citizens' interaction with geospatial Open Government Data (OGD). The system combines semantic data retrieval, agentic reasoning for iterative code generation, and secure sandboxed execution that produces verifiable multimodal outputs. Evaluated on a 199-question benchmark covering both factual and unanswerable questions, across 430 City-of-Zurich datasets and 11 LLMs, OGD4All reaches 98% analytical correctness and 94% recall while reliably rejecting questions unsupported by available data, which minimizes hallucination risks. Statistical robustness tests, as well as expert feedback, show reliability and social relevance. The proposed approach shows how LLMs can provide explainable, multimodal access to public data, advancing trustworthy AI for open governance.
Paper Structure (12 sections, 6 figures, 2 tables)

This paper contains 12 sections, 6 figures, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. METHODOLOGY
  4. Dataset Retrieval
  5. Dataset Analysis
  6. Benchmark
  7. RESULTS
  8. Retrieval
  9. Analysis
  10. Expert Feedback and Deployment Considerations
  11. Limitations and Future Work
  12. CONCLUSIONS

Figures (6)

  • Figure 1: High-level overview of OGD4All, where an initial dataset retrieval stage is followed by a dataset analysis stage.
  • Figure 2: Screenshot of the Web-based prototype showcasing the retrieval output for two questions.
  • Figure 3: Screenshot of the Web-based prototype. The retrieval reformulations, links to utilized datasets, and the coding agent's step-by-step actions can be inspected by opening the accordion UI elements.
  • Figure 4: Benchmark dataset categories & frequent operations
  • Figure 5: Retrieval metrics across different LLMs.
  • ...and 1 more figures