성과공개

4차

구성기술3

경희대학교

학술발표

IFC 건물 정보 모델에서 정확한 엔터티 분류를 위한 BERT 적용

한국CDE학회 2024 동계학술대회 논문집

Syed Haseeb Shah

김인한

BERT, a transformer-based language model, is finetuned for IFC entity categorization, which revolutionizes the classification of BIM building data. This integration enriches BIM data by utilizing BERT's ability to understand complex textual descriptions and metadata related to various building elements using self-attention. The study reveals the complex interactions between different IFC entities, improving the granularity and accuracy of classification. BERT is able to precisely classify and retrieve information by means of exact pre-training on textual data and fine-tuning with IFC-specific information. It is capable of capturing semantic nuances, hierarchies, and relationships embedded within various IFC text. Using it should result in more efficient BIM data processing, improving processes in engineering, construction, and architecture.