Project Requirements Assignment Using BERT-Based Multilabel Classification and Interactive Mapping System

Abstract

A well-developed construction project plan is essential for achieving desired outcomes by ensuring timely completion within budget. When planning construction projects using the design-bid-build (DBB) delivery method, particularly in complex, multiple-division environments such as the Department of Transportation (DOT), clearly assigning project requirements is critical. Defining these requirements clarifies roles and responsibilities, ensures planned implementation, and prevents misunderstandings, errors, conflicts, and collaboration issues that could lead to claims and disputes. The current approach to assigning project requirements primarily depends on the project manager’s manual review of documents to understand requirements and identify relevant divisions based on their experience and judgment, which is time-consuming and prone to errors. This study proposes an automated requirements classification model using an interactive mapping system to address these issues systematically. This study conducts Bidirectional Encoder Representations from Transformers (BERT)-based multilabel classification for putting multiple labels to a project requirement and utilizes a mapping platform system that can address simultaneously the classification results to all divisions. This study conducts a case study with a real-world roadway construction project to evaluate the applicability and validity of the proposed approach. The case study demonstrates the performance of the BERT classifier, with precision scores ranging from 0.81 to 0.91. Additionally, the proposed model enables divisions to swiftly determine their pertinent requirements and identify revisions or updates by accessing the mapping platform. This study contributes an innovative method for reducing conflicts among project participants and ensuring the accuracy of work activity arrangements, thereby enhancing work planning.