Natural Language Processing–Driven Similar Project Determination Using Project Scope Statements

Abstract

Learning from previous projects in light of benchmarking criteria is a desirable and popular approach to reliable project development and planning in the preconstruction phase. Previous similar projects serve as a practical and proven source of knowledge that can be applicable to future projects. In the early preconstruction phase, the common practice of similar project determination involves leveraging simple and limited project characteristics, resulting in determination accuracy degradation. In order to deliver a project context-based similarity evaluation, this study develops and proposes a natural language processing (NLP)–driven method that can recommend similar previous projects by systematically measuring the similarity between project scope statements. NLP techniques enable systematic measurement of project scope similarity, which addresses the reliance on (1) individual experience and expertise for comprehending contents; and (2) time and efforts for reviewing all the unstructured descriptive narratives of project scopes. The proposed method extracts key work activities from project scope statements, evaluates the level of homogeneity (LOH) between extracted activities, and quantifies the project similarity based on the homogeneity evaluation results. The proposed method utilizes bidirectional encoder representations from transformers (BERT) models that can embed unstructured texts into computer-readable numeric formats by considering the context of texts. The output of the proposed model is a graphical map depicting similarities that can help project engineers quickly and intuitively recognize the similarity evaluation results. The validity test shows that the proposed method demonstrates better performance in determining the most highly similar past projects with an ongoing project. The proposed method is appropriate for enhancing an effective information acquisition process from previous projects, resulting in an improved and more efficient project planning process during the preconstruction phase.