Automated Natural Language Building Descriptor for Building Information Models

Abstract: Recent large language models (LLMs) with enhanced multimodal capabilities can interpret drawings and generate building descriptions, yet their capabilities are still limited to simple two-dimensional spatial configurations. This study proposes a method to automatically generate natural language descriptions of spatial programs and the connections between spaces from building information models (BIMs) using threshold-enhanced triangle intersection (TETI) algorithm and reasoning LLMs (OpenAI-o1 and DeepSeek-R1). For validation, the inclusion rate of spaces and the correctness of topological extraction are assessed. Results show that the extracted BIM data effectively capture key features, supporting the potential for natural language-based automated BIM description generation.