Development of potential risk detection model for international construction contracts using natural language processing and machine learning techniques

In international construction projects, many rights claims and disputes arising from construction contracts in which the duties and responsibilities of the parties are determined. The fact that the contract conditions provided by the employer during the tender process of construction projects are not reviewed in detail by the contractor causes these claims and disputes and the parties incur great losses. For this purpose, a model has been developed using natural language processing and machine learning techniques to detect potential risky contract clauses in construction contracts that are in draft form during the tender process. The data set was created using the FIDIC standard contract terms for the developed model. In order to label the created data set, 25 potentially risky clause titles were determined by the survey study and 45 keywords and keyphrases that could be related to the risk were extracted with the keyword extraction application performed on the determined clause titles. In the survey conducted for the extracted keywords, the risk score was determined using the risk matrix and the data set created using these risk scores was labeled. By applying natural language processing techniques to the labeled data set, the data was made ready for the implementation of machine learning algorithms. Six different machine learning algorithms were implemented in the model, and the implemented machine learning algorithms showed successful classification performance and identified potential risky contract clauses. With the study, a fast and efficient technical model was presented to support contract managers in the decision-making process in international construction projects and to be used in the review process of the contract.