Automated construction contract analysis for risk and responsibility assessment using natural language processing and machine learning

Abstract

Construction contracts contain critical risk-related information that requires in-depth examination, yet tight schedules for bidding limit the possibility of comprehensive review of extensive documents manually. This research aims to develop models for automating the review of construction contracts to extract information on risk and responsibility that will provide inputs for risk management plans. Models were trained on 2268 sentences from International Federation of Consulting Engineers templates and tested on an actual construction project contract containing 1217 sentences. A taxonomy classified sentences into Heading, Definition, Obligation, Risk, and Right categories with related parties of Contractor, Employer, and Shared. Twelve models employing diverse Natural Language Processing vectorization techniques and Machine Learning algorithms were implemented and benchmarked based on accuracy and F1 score. Binary classification of sentence types and an ensemble method integrating top models were further applied to improve performance. The best model achieved 89 % accuracy for sentence types and 83 % for related parties, demonstrating the capabilities of automated contract review for identification of risk and responsibilities. Adopting the proposed approach can significantly expedite contract reviews to support risk management activities, bid preparation processes and prevent disputes caused by overlooking risks and responsibilities. • A multi-class taxonomy to label contract sentences into risk and responsibility categories. • Extensive evaluation encompassing around 3500 sentences from FIDIC books and a real contract for validation. • Application of NLP techniques and ML algorithms for text vectorization and classification. • A high accuracy (89 %) was achieved by the ensemble method.

Topics & Concepts