กรอบการแก้ไขอัจฉริยะสำหรับการแปลงเอกสาร PDF เป็นข้อความภาษาไทย

Abstract

This research is an experimental study.  The objectives were to (1) develop a process for converting Thai PDF documents into MS Word documents while preserving the accuracy of the text and document structure, (2) design a framework for correcting Thai text using artificial intelligence, and (3) evaluate the accuracy of the document conversion compared to conventional methods. The samples consisted of Thai PDF documents. The research instrument was the Thai PDF-to-Text Intelligent Correction Framework (TPTICF), which consisted of four steps: extracting data from PDF documents, preparing text, editing text using artificial intelligence, and restructuring the document. The artificial intelligence models Gemini-2.5-Flash, Qwen3-14B, and Typhoon-v2.1-12B-Instruct were compared. Data analysis involved comparing the character correction rate, processing time, system resource usage, and text accuracy based on human evaluation. The research results showed that (1) Gemini-2.5-Flash was the most suitable model for application in the TPTICF framework because it achieved the best balance between text correction, document accuracy, and processing time. (2) Human evaluation of 10 documents with intentionally introduced errors showed that the system had an average accuracy of 99.93% and 7 documents had an accuracy of 100.00% and (3) The TPTICF framework increased the accuracy of converting PDF documents to MS Word documents from an average of 95.08% before AI correction to 99.93% after correction, showing that the developed framework improves the accuracy of converting Thai-language documents.