Automatic Assessment of Leanness of Construction Contract Provisions: A Proof-of-Concept

Automatic Assessment of Leanness of Construction Contract Provisions: A Proof-of-Concept Ramesh Balaji, Sivakumar KS, Murali Jagannathan, Venkata Santosh Kumar Delhi Pages 822-829 (2024 Proceedings of the 41st ISARC, Lille, France, ISBN 978-0-6458322-1-1, ISSN 2413-5844) Abstract: Construction projects significantly contribute to a nation's economic development. However, the sector is synonymous with delays and disputes for various reasons, often due to non-productive work practices. Researchers and practitioners recommend applying lean construction principles to mitigate non-value addition activities and improve productivity and performance. However, existing contract forms may contain provisions that are counter-productive, thereby making lean implementation a challenge. Therefore, when planning to go lean, it becomes important to assess the extent to which a contract provision aids or hinders lean implementation, in other words, 'leanness' assessment. A manual analysis is possible but time-consuming and prone to subjective decision-making. Artificial Intelligence (AI)-backed Language Model (LM)-based tools can be potentially used to quickly and efficiently classify a contract clause based on lean implementation-friendliness. Therefore, a dataset containing 734 contract clauses is manually classified into 14 labels based on the literature review, and a part of this data is used to train Bidirectional Encoder Representations from Transformers (BERT)-LM. With an F1 score of 77%, the study shows that LM-based solutions can be potentially employed for construction contract leanness assessment. The study, which is an initial attempt towards developing a reliable leanness prediction model in the future, also noted that the Bert-base-cased LM performs better than its 'large' counterpart under both the 'cased' and 'uncased' conditions. Keywords: Contracts, Lean, Language Models, Finetuning, Artificial Intelligence DOI: https://doi.org/10.22260/ISARC2024/0107 Download fulltext Download BibTex Download Endnote (RIS) TeX Import to Mendeley