Intelligent Question Answering Technology in the Domain of Architecture Based on Large Language Models

As the scale and complexity of data in the construction industry grow rapidly, traditional methods face significant bottlenecks in improving efficiency and project coordination. To address this issue, an innovative construction-specific question-answering system, ‘ArchGPT,’ based on a large language model (LLM) is proposed in this paper, aiming to tackle the two core challenges: high-quality domain data construction and the hallucination issue of large models. The paper proposes a dataset construction approach based on a diversified construction and optimization strategy, and successfully constructs a large-scale, high-quality construction-specific question-answer dataset, which provides a reliable data foundation for system development. At the implementation level, this paper utilizes the LangChain framework to enable seamless integration of the generative model and domain knowledge base. Through prompt learning techniques, the hallucination issue is mitigated, significantly enhancing response accuracy and reliability. The experimental evaluation demonstrates that the dataset constructed in this paper exhibits significant advantages in knowledge integration, citation accuracy, and domain relevance in the field of architecture. Furthermore, compared to systems driven by general LLMs, the ArchGPT system shows significant improvements in key metrics, including accuracy, logical consistency, and domain-specific performance, thereby offering robust technical support for the digital transformation and intelligent upgrading of the construction industry.