Latent Dirichlet Allocation-Based Approach for Automatically Mapping Components to Tasks in Modular Construction

A large portion of cross-knowledge domain tasks have interdependent relationships with varied components in modular construction (MC). The MC components serve as the critical resources to support the task planning and execution for generating excellent MC products and services. Meanwhile, dynamic changes of tasks may adversely affect the design, procurement, and assembly of components. Furthermore, manually mapping components to tasks will be time-consuming and prevent forming effective work packages to achieve collaborative working. Thus, this study aims to develop an approach for automatically connecting components with tasks, which helps workers efficiently know the relationships between tasks and components. To this end, the latent Dirichlet allocation (LDA) approach is customized to this task-component mapping scenario. Moreover, compared with other leading unsupervised clustering techniques, e.g., K-means, the customized LDA demonstrated better performance on accuracy and efficiency for task-component mapping, and it can pave the way for effective work package formation in MC.