Article Details

Abstract

To address the issues of inconsistent naming conventions and high retrieval thresholds in Building Information Modeling (BIM) attributes, this paper designs a graph-enhanced retrieval generation system (Graph-RAG) using the DeepSeek-R1 large language model and the Neo4j graph database. The system employs a regular expression-enhanced natural language to Cypher query (NL2Cypher) mapping technique and a logic self-healing interceptor to enable an automated closed-loop workflow from Industry Foundation Classes (IFC) models to engineering audit reports. Experimental results show that the response time during the mapping phase remains below 812 milliseconds. In a test involving 230 components with missing material attributes, the system achieves a 100% instruction success rate through latent semantic mining, completing the task in 61.151 seconds. The study demonstrates that the system effectively overcomes the reliance on standardized modeling data, significantly improves BIM data governance efficiency, and enhances decision-making support, offering a novel approach to the digitalization of engineering audits.

Keywords

BIM; DeepSeek-R1 large language model; knowledge graph; logic self-healing; Graph-RAG; NL2Cypher; engineering audit; attribute flattening

References

1. [1] Chen Yi'an, Jiang Huixian Optimization Strategy of Node Feature Enhancement Graph Neural Network for BIM Spatial Recognition and Classification [J]. Journal of Earth Information Science, 2025, 27 (12): 2802-2817
2. [2] Li Liang, Ma Zhiliang, Zhu Honggang, etc A Review of Research on Building Information Modeling Standards [J]. Science, Technology and Engineering, 2025, 25 (20): 8347-8363
3. [3] Liu Meiyu, Liu Xuesong, Yang Wuchuan, etc Design and Research of IT Technology Learning Knowledge Graph System Based on Neo4j [J]. Modern Information Technology, 2025, 9 (16): 113-119
4. [4] Liu Yixin, Zheng Ying, Zhang Shuangshi, etc Research on the Construction of Social Stability Risk Knowledge Graph [J]. Modern Information Technology, 2025, 9 (11): 131-138143
5. [5] Zou Fubing, Liu Fan, Liu Qiang, etc Dynamic Graph Query Method Based on Top-k Continuous Matching Mechanism [J]. Computer Applications and Software, 2025, 42 (2): 343-354
6. [6] Wu Zhixiang, Sha Huanxu, Yin Lulu, etc Research on Academic and Social Impact Q&A of Renowned Scientists Based on Graph RAG Semantic Fusion [J]. Intelligence Theory and Practice, 2026, 49 (3): 160-169
7. [7] Ren Haiyu, Liu Jianping, Wang Jian, etc A review of research on intelligent question answering systems based on big language models [J]. Computer Engineering and Applications, 2025, 61 (7): 1-24
8. [8] Zeng Fanshuang, Zhang Naiwen, Li Yingzhuo, etc Ground small target detection system based on YOLOv8 model and Streamlit [J]. Modern Information Technology, 2025, 9 (19): 57-61, 69
9. [9] Wang Qiushi, Qin Tianzhu, Han Kaining, etc Sparse Code Division Multiple Access Technology for Semantic Communication [J]. Signal Processing, 2025, 41 (10): 1614-1623
10. [10] Kong Guoliang, Hu Xuhong, Zheng Zhilin, etc Research and development of railway intelligent route selection system based on cost graph path planning [J]. Journal of Railway Engineering, 2024, 41 (10): 18-23

Funding

Jiangsu Province Education Science Planning Project (C/2024/02/04); Jiangsu Vocational Education Research Project of 2025-2026 by Jiangsu Vocational Education Association (XHYBLX2025284); Research Project on Informationization of Higher Education in Jiangsu Province (2025JSETKT230); General Project of Philosophy and Social Sciences Research in Jiangsu Universities (2025SJYB1587)

About the authors

Zhao Jinlei (1988-), male, Han ethnicity, from Linyi, Shandong, is a senior engineer with a PhD. His research focuses on the application of artificial intelligence technology in engineering.