Research on a Dual-Source Aerodynamics and Flight Principles Knowledge Q&A System Based on GraphRAG

Abstract

Addressing the challenges of outdated textbook knowledge and fragmented internet information within the aviation domain, this paper proposes a dual-source aerodynamics and flight principles knowledge Q&A system based on GraphRAG. This system integrates the authoritative rigor of textbooks with the timeliness of internet information. A knowledge graph is constructed from self-compiled textbooks, supplemented by an internet timeliness knowledge anchoring mechanism, forming a cohesive dual-source knowledge framework. A domain-enhanced GraphRAG strategy is designed to significantly improve the accuracy and comprehensiveness of retrieved results. A knowledge fusion mechanism is implemented to facilitate multi-source knowledge conflict. By fine-tuning the DeepSeek, the system generates systematic, authoritative answers tailored to users' query. The prototype system shows that it outperforms traditional aviation Q&A systems, with an average improvement of 24.8%. The proposed architecture enables migration to other domains by replacing domain knowledge sources.