Automated government document analysis and classification using robotic process automation-ChatGPT model

Abstract

Government offices still process most official documents manually, and the cost is familiar: slow turnaround, inconsistent filing, and data-entry mistakes that are easy to make when every document must be read, classified, and re-keyed. This study tackles that problem with an automated document-processing system for the University of Phayao Document Management System (UP-DMS) that brings together robotic process automation (RPA), optical character recognition (OCR), and the chat generative pre-trained transformer application programming interface (ChatGPT API). A software bot retrieves each document, OCR pulls out the text, ChatGPT classifies the document and summarizes its contents, and the bot writes the result back into UP-DMS. On the OCR pilot, the system reached 89.00% accuracy, enough to support the later stages reliably. Tested on 105 authentic administrative documents, it classified the three main document types at 95.24% accuracy (macro-F1 = 0.952) and the fifteen sub-categories at 82.86%. Automation also cut the average handling time from roughly 4.5 minutes per document to 0.38 minutes, a 91.57% reduction. When 24 administrative staff trialed the system, they rated it highly for usability (x̄ = 4.49), performance (x̄ = 4.49), and practical impact (x̄ = 4.60). The findings point to a workable way of reducing workload while improving both the speed and the accuracy of document handling in public and academic institutions.