Embodied-State-Driven Agentic GraphRAG for Contextual Interpretation of Ancient Chinese Alchemical Heritage
Year: 2026
Type: preprint
Abstract: Ancient Chinese alchemy is an important part of scientific and technological heritage, preserving historical knowledge of materials, apparatus and technical practices. However, fragmented sources and largely scripted digital interpretations make it difficult to connect these practices with reliable evidence and modern scientific understanding during interaction. We developed an extended reality (XR) system integrating a provenance-aware knowledge graph with Agentic GraphRAG. Its Embodied State as Query mechanism uses process stages, objects, actions and material states to retrieve historical passages and contextual scientific explanations. Agentic GraphRAG achieved the highest scores across four offline retrieval and interpretation measures. In a study of 66 participants, adaptive XR yielded greater conceptual and procedural knowledge gains than web-based learning. Compared with scripted XR, it received higher ratings for explanation credibility and source clarity and elicited more mechanism-oriented questions. These findings support state-conditioned, evidence-grounded interpretation of historical technical practices during interaction.
Source: Research Square
Authors: Yumiao Chen, Tanghua Xie
Institution: East China University of Science and Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7236228510
