Research on Constructing Immersive Learning Environments for Cultural Heritage Based on Retrieval-Augmented Generation and Intelligent Agents

Abstract

In order to address the challenge of two-dimensional teaching resources in supporting spatial, structural, and cultural depth cognition in Chinese cultural heritage education, this study takes the East Hall of Foguang Temple as the research object and designs and implements a cultural heritage immersive intelligent learning environment based on enhanced RAG and intelligent agents. This architecture mainly includes three key technological contributions: firstly, a context aware retrieval algorithm that integrates virtual spatial coordinates, visual focus, and semantic queries is designed, achieving accurate mapping from low-level interaction signals to high-level semantic queries; Secondly, a retrieval result filtering and reordering model based on domain knowledge graph was constructed, which effectively improved the retrieval accuracy and relevance under complex professional problems by calculating semantic relevance and graph path distance; Finally, a constrained large language model inference and structured output control mechanism was implemented, which ensured the academic rigor and traceability of generated content in free dialogue mode through dynamic prompt engineering and post-processing validation. This architecture achieves a self-adaptive teaching loop of “questioning in virtual scenes, evidence-based reasoning responses, and guidance for subsequent learning” through knowledge graph based retrieval filtering and context aware prompting engineering. A two-week randomized controlled experiment involving 60 new architecture students showed that compared to the traditional teaching group, the experimental group using this system exhibited significant advantages in comprehensive understanding, spatial structure cognition, and learning participation. This work not only provides a verifiable intelligent solution for cultural heritage education, but also provides specific implementation examples and performance verification in computer application technologies such as cross modal information retrieval, large model interaction for domain knowledge enhancement, and immersive learning environment construction.