Spelling Correction Non-Word Error pada Dokumen Bahasa Indonesia dengan Pendekatan Fasttext

Abstract

Technological advancements have influenced document writing, where critical issues such as typographical errors or spelling mistakes are commonly found. Proper document writing relies heavily on the use of correct spelling that adheres to the official language guidelines outlined in the Ejaan Yang Disempurnakan (EYD) based on the Kamus Besar Bahasa Indonesia (KBBI). Therefore, a system capable of checking and automatically correcting spelling errors is essential to address such issues in a document. Most existing spell correction systems are limited to processing a few sentences and require significant processing time. This study develops a spell correction system that implements the FastText approach and Peter Norvig's method, designed for Android-based platforms. A pre-trained FastText model is utilized to generate word candidates in the application of Peter Norvig's method. The Kamus Besar Bahasa Indonesia (KBBI) is employed as the word dataset in this research. The testing results of the developed spell correction system achieved a recall of 79.57%, precision of 91.79%, accuracy of 91.05%, and an F1 score of 85.24%. The system accepts input in the form of Indonesian documents with file extensions Microsoft Word (.docx) and PDF (.pdf).