Automated Construction Contract Summarization Using Natural Language Processing and Deep Learning

Abstract

Automated Construction Contract Summarization Using Natural Language Processing and Deep Learning Xiaorui Xue, Yiru Hou and Jiansong Zhang Pages 459-466 (2022 Proceedings of the 39th ISARC, Bogotá, Colombia, ISBN 978-952-69524-2-0, ISSN 2413-5844) Abstract: The interpretation of construction contracts is crucial to the management and success of a project. Correct and accurate interpretation could support the smooth construction of high-quality built assets. Misunderstanding and omissions may lead to costly rework and delay. One main challenge of construction contract interpretation lies in the length of construction contracts. Therefore, the demand for reducing the length of such documents while keeping their main information elements emerges. To address this research need, the authors proposed to use natural language processing (NLP) and deep learning technology to summarize construction contracts (i.e., text summarization). There are many deep learning models available and developed for text summarization. However, their performance on construction contracts is to be tested. To address this gap, the authors proposed a new merit-based evaluation method to evaluate the performance of three deep learning models on text summarization of construction contracts, which were reported the state-of-the-art performance on text summarization tasks in general English corpus. The proposed method evaluated selected models from three aspects: information completeness, information correctness, and human readability. The Distilbart model, which scored 5.23, 4.82, and 5.05 in these three aspects, respectively, outperformed the other models in all three aspects. Keywords: Deep Learning; Automated Construction Contract Summarization; Text Summarization; Natural Language Processing; Artificial Intelligence; Construction Management DOI: https://doi.org/10.22260/ISARC2022/0063 Download fulltext Download BibTex Download Endnote (RIS) TeX Import to Mendeley