Multi-Class Categorization of Design-Build Contract Requirements Using Text Mining and Natural Language Processing Techniques
Fahad Hassan, Tuyen Le, Duc-Hoc Tran
Abstract
Contract requirement processing is one of the most crucial tasks in the construction projects practicing design-build project delivery method. The contract requirements and technical specifications of design-build projects are typically enormous and associated with many different disciplines and project stages. In many cases, the design-build contractor needs to classify these general contract requirements into different categories to prepare the subcontracts for the specialized works. The success of a design-build project highly relies on the completeness of subcontracts to avoid disputes which may lead to delays and cost overruns. The conventional practices for the preparation of subcontracts require professionals much time and effort to read the complete contract package and extract the requirements related to the specialized works. This paper introduces an effective method to prepare the subcontracts scope by developing an automated framework for information retrieval using natural language processing techniques. The proposed technique classifies the text describing project requirements into three distinct classes associated with different construction project stages namely as design, construction, and operation and maintenance. The requirement classification model was developed using six different supervised machine learning approaches, including naïve Bayes, support vector machine, logistic regression, K-nearest number, decision tree, and feedforward neural network. The experimental results revealed the logistic regression as the highest performing algorithm with an accuracy of 94.12%. This study is expected to help reduce reading time and improve the quality of subcontracts.