Automatic Risks Detection and Comparison Techniques for General Conditions of Technical Documents in Purchasing Order

Abstract

This research is to develop a technique that recognizes the technical documents as part of purchasing order (PO) exchanged between the owner (buyer) and the supplier (seller) in capital investment, such as maintenance and replacement of equipment, automatically detect specific potential risk contained clauses and shows the comparison results. This research has selected the proof of concept (PoC) technology to (1) the performance guarantee clauses for the purchasing equipment and (2) the delivery schedule requirement clauses to be checked and compared with the utmost cares when reviewing technical documents by the plant owner. The PoC research was implemented based on the Python programming language in conjunction with the spaCy libraries, and further was developed to a cloud-platform-based application for user implementation. This technique preprocesses technical documents of PO in PDF format and, after recognizing and converts into the entire text, detects and extracts the risks-related sentences with logic created by analyzing the patterns of PoC clauses. This research also built a database of all units and formats that can be used in PoC clauses and developed knowledge-based rules to normalize PoC clauses expressed differently in two (buyer's and seller's) and documents. Finally, the result of comparing PoC clauses unified in the same unit and format is output to an Excel or CSV file. Also, these techniques and comparison results were verified through the confusion matrix and accuracy-check. This study is expected to reduce the workload and improve practitioners' productivity in engineering procurement processes for capital investment projects.