Natural Language Processing–Driven Model to Extract Contract Change Reasons and Altered Work Items for Advanced Retrieval of Change Orders

Abstract

Change orders are documents that describe a specific contract amendment to the original scope of work. Historical change orders are invaluable information sources that can provide practical and proven solutions for developing new change orders from similar cases. However, current change order management systems are not efficient in searching for and finding the most related and similar change orders due to inherent weaknesses in current archiving and search processes, such as keyword-based or reason code–based search. This study proposes and develops a natural language processing (NLP)–driven model that can significantly improve the accuracy and reliability of searching cases by restructuring how each change order’s information is stored and retrieved in change order management systems. The NLP-driven model proposed in this study can automatically detect change reasons and altered work items through text representation pattern analysis and training. The proposed model applies semantic frames to define essential semantic components and determines syntactic features for text representation pattern analysis. The model also utilizes a conditional random field (CRF) classifier, which can consider contexts in sequential texts at the model training stage. The proposed model can significantly improve the accuracy and relevancy of the search process to find the most similar cases by allowing context-driven classification, archiving, and retrieval of change orders.