A Natural Language Processing-Based Approach for Clustering Construction Projects

Many construction project owners group their projects into different work types to facilitate effective project management decisions. This categorization significantly helps owner agencies narrow down and analyze the historical projects of a similar type to extract meaningful patterns that can support various project management decisions. However, many owners, particularly state highway agencies, typically rely on project engineers' subjective judgments to classify a new project or do not even have systematic project classification criteria. In addition, many projects are a mixture of work types with varying proportions. A systematic and objective process of classifying projects is desirable to generate a more accurate and less disputable categorization of projects, thereby improving project management decision-making. This study proposes a natural language processing-based model for grouping projects with similar work components and portions into the same group. For a specific project, work items' descriptions and cost composition are the key input variables of the model. Bid tabulation data from a highway agency were collected and used for model development and evaluation.