Unlocking the power of language models for smart configuration in the AEC domain using spoken language understanding

The convergence of Industry 4.0 demands and the rapid advancement of artificial intelligence technologies presents unprecedented opportunities for the architecture, engineering, and construction (AEC) industry. Natural Language Processing (NLP), the flagship technology of artificial intelligence that focuses on understanding and processing human language, offers numerous tailor-made solutions for smart design and construction in the AEC domain. Among them, one crucial aspect is the characterization of user needs, which typically involves transforming unstructured user texts into structured information, facilitating mass customization within the context of smart configuration. However, current approaches to understanding user needs usually rely on heuristic or traditional machine learning methods, resulting in significant errors and limited accuracies. This chapter introduces a novel perspective on smart configuration for mass customization by incorporating language models into user needs understanding. Specifically, we began by framing the user needs understanding task as a spoken language understanding task, addressing practical challenges through concrete approaches and experimental designs. In this process, we identified two key roles for language models in smart configuration: 1) serving as knowledge bases and 2) functioning as backbone models. Additionally, we discussed the opportunities and challenges associated with employing language models in this domain.