Intelligent question and answer system for building information modeling and artificial intelligence of things based on the bidirectional encoder representations from transformers model

Tzu-Hsuan Lin,Yu-Hua Huang,Alan Putranto
DOI: https://doi.org/10.1016/j.autcon.2022.104483
IF: 10.3
2022-07-20
Automation in Construction
Abstract:In recent years, building information modeling and artificial intelligence of things (BIM-AIOTs) in the construction industry have gained much attention. Construction engineers and researchers learn about BIM-AIOT and increase their professional knowledge through internet searches. However, the large amount of information on the internet makes it difficult to find specific information. Although some previous work of BIM-related searches exists, most still search with a combination of keywords or longer terms. This paper utilizes a machine learning model with natural language processing (NLP) technique of bidirectional encoder representations from transformers (BERT) integrated with a mobile chatbot as a question and answer (QnA) system. The dataset used for modeling contained 3334 text paragraphs that shortened to 10,002 questions. The result shows an F1 score of around 65% accuracy, which is acceptable for model prediction. Then, the system verifies to synchronize to the server and user interface. The system works well for information search and offers a supporting automation information system in the construction industry. This study achieved conversational machine understanding and a user-friendly BIM-AIOT integration information searches platform. The proposed system has a reliable research-based information source. It is verified as an effective and efficient way to produce fast decision-making. The system is deemed a future application for research-based problem-solving solutions in Architecture, Engineering, and Construction (AEC).
What problem does this paper attempt to address?