Intelligent Question-Answering on Geomorphology Knowledge Based on Knowledge Graph Retrieval-Augmented Generation Technology

Geomorphology is a discipline dedicated to the study of the morphological characteristics, genesis, distribution, and evolutionary patterns of the Earth's surface. Its intrinsic disciplinary attributes and research domains underscore its pivotal foundational role within the field of geography. However, general-purpose large language models lack training in vertical domains, resulting in suboptimal performance in the field of geomorphology-related question answering. By leveraging knowledge graph retrieval-augmented generation technology, we construct knowledge graphs and knowledge graph communities. Through the integration of graph structures and external knowledge bases, we enhance the responses of large language models, achieving a deep fusion between large language models and knowledge graphs. Using karst landform knowledge as an experimental case for validation, we employed a comparative analysis approach to evaluate the question-answering performance of large language models based on knowledge graphs from both subjective and objective dimensions. The results indicate that, compared to traditional retrieval-augmented generation, knowledge graph retrieval-augmented generation technology demonstrates improvements in the logicality, depth of knowledge, and interpretability of responses. This advancement provides a novel tool for knowledge discovery and reasoning in geomorphological research.