Client-centered detached modular housing: natural language processing-enabled design recommender system

Abstract Designing modular housing is a complex task that necessitates a thorough understanding of the diverse needs of clients in terms of both design aesthetics and floor plan layout. Furthermore, adhering to design for manufacture and assembly (DfMA) principles adds to the complexity, as these are essential modular home requirements. Traditional construction methods frequently fail to meet the specific needs of both clients and DfMA, potentially resulting in suboptimal design solutions. Incorporating client requirements during the design phase necessitates the use of an effective system and framework to reduce changes in subsequent project stages. Existing literature lacks a suitable approach, particularly in the context of modular housing. To address this gap, this paper introduces an artificial intelligence–building information modeling recommender system (RS) for detached modular housing design. The system processes client requirements entered as text utilizing the Word2vec algorithm with the GloVe dataset, refined through transfer learning using surveyed client data of housing needs. The system recommends three distinct modular building design alternatives sourced from a building information modeling models database using cosine and Euclidean similarity functions. A sensitivity analysis ensures that client needs are considered fairly, increasing the robustness of the RS. By incorporating natural language processing, this system transforms the construction industry by making initial designs more client-centric compared with traditional methods. Furthermore, it promotes improved collaboration among clients, design, and construction teams, reducing modifications to design in later stages of construction.