Domain-Specific Language Model for intelligent construction contract analysis

Construction projects frequently encounter claims and disputes due to misinterpretation of complex contractual provisions, while manual review remains time-consuming and inconsistent, reducing the effectiveness of contract administration. This study proposes an intelligent framework for construction contract analysis using fine-tuned Domain-Specific Language Models (DSLMs). The framework automatically extracts and classifies contractual obligations and entitlements of the Employer, Contractor, and Engineer. Seven DSLMs derived from Large Language Models (LLMs) with varying architectures and parameter sizes are evaluated and cross-validated on multiple standard forms and bespoke/ad hoc contracts. The best-performing model achieves 98.86% accuracy, outperforming BERT-based and classical machine learning baselines. Results demonstrate that fine-tuned DSLMs enhance language models' capability to interpret and classify contractual provisions accurately. The proposed framework provides practitioners with a tool to accelerate contract review, improve consistency in identifying obligations and entitlements, reduce manual effort, and mitigate the risk of disputes arising from contractual misinterpretation. It supports efficient contract administration.