[Objectives] The trajectory knowledge graph effectively captures the deep semantic relationships between trajectories and geospatial entities, offering significant advantages in revealing complex associated information. However, traditional methods for constructing knowledge graphs from domain-specific data sources rely heavily on expert knowledge, involve extensive data preprocessing and entity-relationship extraction, and require high levels of professional expertise. [Methods] To address these challenges, this paper proposes a trajectory knowledge graph construction method that supports natural language-driven task execution through prompt learning with large language models. First, a prompt strategy for the preprocessing task is designed to guide large language models in automatically generating data processing code for cleaning abnormal trajectories. Second, a two-level system prompt strategy is developed to enable tool invocation by matching and calling the trajectory knowledge extraction tool. This strategy allows non-expert users to complete the graph construction process using simple natural language instructions, significantly reducing reliance on programming skills and deep semantic understanding. [Results] To evaluate the feasibility and effectiveness of the proposed prompt strategies, a set of test sentences was created for trajectory preprocessing and entity-relation extraction tasks. Real-world ship and vehicle trajectory datasets were used to support knowledge graph construction. Experiments conducted on two representative large language models, Tongyi Qianwen and Baidu Qianfan, achieved average accuracy rates exceeding 75% and 80%, respectively, demonstrating strong generalization ability and practical value. [Conclusions] This study verifies the effectiveness of combining large language models with prompt learning in constructing trajectory knowledge graphs with low technical barriers, demonstrating the strong generalization and application value of the proposed prompt strategy.

[Objectives] The deep integration of Graph Neural Network (GNN) and Building Information Modeling (BIM) has significantly advanced the Architecture, Engineering, and Construction (AEC) industry. In this paper, we propose a self-supervised GNN model that enhances node features to address the spatial identification and classification problem in BIM, thereby closing the gap between high‑fidelity BIM data and graph‑based representation learning through a task‑oriented framework that captures both geometric and functional semantics. [Methods] The model represents each building space as a node and fuses edge features based on node characteristics. To maximize the descriptive power of the features, topologically meaningful relationships—shared‑wall, passageway, corner, and open‑space connections—are encoded as weighted edges and injected into the node embeddings through a learnable fusion layer before graph propagation. The model is trained on an autonomously constructed knowledge base that includes 12 space types and 4 relationship features across 3 major types of building-space layout graphs, enabling the automatic recognition of functional space types in buildings. [Results] The experimental results show that (1) on the BuildingGraph benchmark, which spans school, apartment, and office typologies and comprises 300 large‑scale architectural layout subgraphs, the enhanced GNN with fused edge features achieved accurate building‑space node classification, attaining 96.83% accuracy while maintaining stable convergence of the training loss on graphs containing thousands of nodes. (2) Compared with existing GNN models under identical hyper‑parameter settings in ablation experiments, BGFEF significantly improved performance in terms of accuracy (97.08%) and F₁ score (96.75%), surpassing Graph‑BERT and SAGE‑E by 3%~12% (3) In real‑world applications, nine BIM models across school, apartment, and office buildings were evaluated by extracting room connectivity topology graphs, and the method's spatial classification accuracy and interoperability were verified, with the best single school instance reaching 96.87% accuracy. [Conclusions] This study improves the efficiency and automation of spatial analysis and provides practical guidance for implementation in real AEC projects. By producing high‑confidence functional labels that map seamlessly to Industry Foundation Classes (IFC), the proposed model pipeline accelerates rule‑based design review, facility management, and evacuation‑route optimization, offering a scalable foundation for future ontology‑grounded graph reasoning across diverse building typologies.