Comparing Natural Language Processing Methods to Cluster Construction Schedules

The names of construction activities are the only unstructured data attribute in schedules, and they often guide execution. Activity devised to communicate between stakeholders, therefore written using inconsistent terminologies across repetitive with omitted contextual information. This presents a challenge for machine learning systems when patterns from schedules. paper compared performance state-of-the-art text-related clustering methods identifying activities. was achieved by creating ground truth set on basis standard work classification, then comparing precision, recall, F1 score latent semantic analysis (LSA), Dirichlet allocation (LDA), word2vec, fastText algorithms group activity 27 Results indicated that LSA outperformed LDA (0.84% versus 0.88%), whereas results language models–based depended quality word embedding paired method. study provides insight into how preprocess schedules further artificial intelligence (AI)-based quantitative analysis. Methodologies described this will help researchers who natural language–related research (e.g., safety contract management) better capture feature words, rather than counting frequencies.