Enhancing Geographic Inquiry through Cooperative Integration of Large Language Models and Domain-Specific Knowledge Bases

Keywords: Geography Information Science, Large Language Model
Abstract Type: Paper Abstract

Abstract

The rapid development of large language models (LLMs) offers transformative potential for geographic research, which demands interdisciplinary insights and domain-specific expertise. Despite their potential, general-purpose LLMs face limitations in addressing geography-specific questions. These models can generate broadly acceptable responses, yet often lack the reliability and depth needed for nuanced geographic analysis, making them insufficient for the domain’s rigorous demands. To address these gaps, this study explores a cooperative model that integrates LLMs with a structured body of knowledge in geography. This collaboration aims to balance the generative flexibility of LLMs with the precision of a domain-specific knowledge base, enhancing both accuracy and contextual relevance in geographic question answering. This study addresses these challenges by pursuing three interconnected objectives: (1) identifying and analyzing the specific limitations of general LLMs within the geography domain, highlighting gaps in accuracy and contextual depth; (2) developing a benchmark dataset specifically designed for metric-based geographic question answering, providing a standardized means to assess LLM performance on domain-relevant tasks; and (3) proposing a novel approach to constructing a geography-specific LLM that integrates a structured body of knowledge database with the generative capabilities of LLMs, aiming to enhance response accuracy and domain relevance.