Mitigating Geospatial Knowledge Hallucination in Large Language Models: Benchmarking and Dynamic Factuality Aligning

TL;DR

This paper introduces a benchmark and a dynamic alignment method to evaluate and reduce geospatial hallucinations in large language models, significantly improving their factual accuracy in geospatial tasks.

Contribution

It presents the first systematic framework for evaluating and mitigating geospatial hallucinations in LLMs using structured knowledge graphs and a novel optimization technique.

Findings

Over 29.6% performance improvement on the benchmark

Effective reduction of geospatial hallucinations in LLMs

Enhanced trustworthiness in geospatial reasoning tasks

Abstract

Large language models (LLMs) possess extensive world knowledge, including geospatial knowledge, which has been successfully applied to various geospatial tasks such as mobility prediction and social indicator prediction. However, LLMs often generate inaccurate geospatial knowledge, leading to geospatial hallucinations (incorrect or inconsistent representations of geospatial information) that compromise their reliability. While the phenomenon of general knowledge hallucination in LLMs has been widely studied, the systematic evaluation and mitigation of geospatial hallucinations remain largely unexplored. To address this gap, we propose a comprehensive evaluation framework for geospatial hallucinations, leveraging structured geospatial knowledge graphs for controlled assessment. Through extensive evaluation across 20 advanced LLMs, we uncover the hallucinations in their geospatial knowledge. Building on these insights, we introduce a dynamic factuality aligning method based on Kahneman-Tversky Optimization (KTO) to mitigate geospatial hallucinations in LLMs, leading to a performance improvement of over 29.6% on the proposed benchmark. Extensive experimental results demonstrate the effectiveness of our benchmark and learning algorithm in enhancing the trustworthiness of LLMs in geospatial knowledge and reasoning tasks.