Contractor’s Risk Analysis of Engineering Procurement and Construction (EPC) Contracts Using Ontological Semantic Model and Bi-Long Short-Term Memory (LSTM) Technology

Abstract

The development of intelligent information technology in the era of the fourth industrial revolution requires the EPC (engineering, procurement, and construction) industry to increase productivity through a digital transformation. This study aims to automatically analyze the critical risk clauses in the invitation to bid (ITB) at the bidding stage to strengthen their competitiveness for the EPC contractors. To this end, we developed an automated analysis technology that effectively analyzes a large amount of ITB documents in a short time by applying natural language processing (NLP) and bi-directional long short-term memory (bi-LSTM) algorithms. This study proposes two models. First, the semantic analysis (SA) model is a rule-based approach that applies NLP to extract key risk clauses. Second, the risk level ranking (RLR) model is a train-based approach that ranks the risk impact for each clause by applying bi-LSTM. After developing and training an artificial intelligent (AI)-based ITB analysis model, its performance was evaluated through the actual project data. As a result of validation, the SA model showed an F1 score of 86.4 percent, and the RLR model showed an accuracy of 46.8 percent. The RLR model displayed relatively low performance because the ITB used in the evaluation test included the contract clauses that did not exist in the training dataset. Therefore, this study illustrated that the rule-based approach performed superior to the training-based method. The authors suggest that EPC contractors should apply both the SA and RLR modes in the ITB analysis, as one supplements the other. The two models were embedded in the Engineering Machine-learning Automation Platform (EMAP), a cloud-based platform developed by the authors. Rapid analysis through applying both the rule-based and AI-based automatic ITB analysis technology can contribute to securing timeliness for risk response and supplement possible human mistakes in the bidding stage.