Auto-Summarization for the Texts of Construction Dispute Precedents

For the effective construction dispute management, it is important to quickly identify the similar precedent cases as it helps practitioners make decision for the direction to respond appropriately to disputed issues. However, due to their extensive length and specialized content, promptly and accurately comprehending texts of precedent cases related to construction disputes poses challenges. This study presents a model that employs natural language processing (NLP) for automatically summarizing texts of precedent cases relevant to construction disputes. Developed using Python, this tool generates summary reports for 300 US construction dispute precedents obtained from the Westlaw database. During the preprocessing phase, case texts undergo processing based on construction knowledge. To ascertain the most suitable model for summarizing construction dispute texts, various summary models utilizing Bidirectional Encoder Representations from Transformers (BERT) and traditional ranking algorithms like TextRank and RexRank are compared. The efficacy of these summarized outcomes is evaluated not only using the general ROUGE algorithm but also through readability assessments by domain experts. The findings of this research have the potential to aid practitioners in the timely management of dispute-related documents by automating the summarization process.